1"use strict";(self.webpackChunkNioLandReact=self.webpackChunkNioLandReact||[]).push([[3624],{3624:(e,a,s)=>{s.r(a),s.d(a,{default:()=>h});var t=s(5043),i=s(5475),n=s(1072),r=s(8602),l=s(3107),o=s(7497),d=s(579);function c(){const[e,a]=(0,t.useState)(0),[s,l]=(0,t.useState)(0),[o,c]=(0,t.useState)(-1);function h(s){a(e===s?-1:s),m(-1),function(e){l(e)}(s)}function m(e){c(e)}const p=[{title:"Demystifying AI",firstPage:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("img",{src:"images/help-center/aibanner.webp",alt:"cover",loading:"lazy"}),(0,d.jsx)("h3",{className:"h5"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:'These days, everything seems to have "AI" attached to it. Our journey with AI began back in 2013, when the buzzword was for AI was still "big data." Since then, advances in algorithms, processing power have dramatically expanded AI\'s capabilities. However, key factors like data cleaning, processing, and measuring success are still often overlooked in conversations about AI. '}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"If you're considering starting an AI pilot project or deploying the digital supply chain, it\u2019s essential to understand what makes AI effective. In this section, we try and simplify the concepts and processes behind AI and machine learning (ML) to provide you with starting point to understand them."}),(0,d.jsx)("h3",{className:"h5",children:"What is AI ?"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Artificial Intelligence (AI) refers to systems designed to perform tasks that typically require human intelligence, such as recognising speech, making decisions, or predicting outcomes. AI is often categorised into:"}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"Narrow AI: "})," Focused on specific tasks, like facial recognition or automated translations."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"General AI (hypothetical):"})," A system capable of reasoning and solving problems across any domain, which remains a goal for the future."]}),(0,d.jsx)("h4",{className:"h6",children:"AI is powered by Machine Learning (ML)"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"At its core, AI is powered by Machine Learning (ML), a subset of AI that enables systems to learn from data rather than following explicitly programmed rules."}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("img",{src:"images/help-center/supervised.webp",alt:"i",style:{width:"2.5rem",marginRight:"0.5rem"},loading:"lazy"})," Supervised Learning: Algorithms learn from labelled data, akin to teaching a student with answers provided as guidance."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("img",{src:"images/help-center/unsupervised.webp",alt:"i",style:{width:"2.5rem",marginRight:"0.5rem"},loading:"lazy"})," Unsupervised Learning: Where the system is left to find patterns and connections without labeled data , without guidance."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("img",{src:"images/help-center/reward.webp",alt:"i",style:{width:"2.5rem",marginRight:"0.5rem"},loading:"lazy"})," Reinforcement Learning: Where the system is left to find patterns and connections without labeled data , without guidance."]}),(0,d.jsx)("h3",{className:"h5",children:"Tackling Complex Challenges with AI and ML"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:'For more complex problems, AI systems often use deep learning, a method that relies on neural networks inspired by the human brain. These networks are designed with interconnected layers of nodes ("neurons") to process information and identify intricate patterns. While these networks draw inspiration from biology, they are highly simplified mathematical models, far less complex than the human brain.'}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Deep learning is particularly effective for tasks like image analysis, speech recognition, and uncovering hidden trends in vast datasets. These advancements have made AI and ML indispensable tools for transforming raw data into actionable insights."}),(0,d.jsx)("h3",{className:"h5",children:" Why It Matters"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"AI and ML are not just about advanced technology\u2014they are practical tools for navigating challenges in today\u2019s complex world. By harnessing these technologies, businesses can improve decision-making, optimise operations, and drive innovation. From weather forecasting to supply chain management, the potential applications of AI are as varied as they are impactful."})]})})}),content:[{title:"What is Machine Learning?",page:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("h3",{className:"h5",children:"What is Machine Learning (A basic defination)"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Machine Learning (ML) is a powerful approach within AI that enables computers to learn and improve from experience. Rather than being explicitly programmed for every task, ML algorithms analyse data, identify patterns, and adjust their behaviour based on what they observe. Over time, this process allows systems to improve their predictions or decisions as they process more data."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"ML is particularly valuable because it adapts and evolves, creating accurate models that adju
1st as new information is introduced. For instance, a spam detection system can continuously refine its ability to filter emails by learning from past examples. This flexibility makes ML indispensable for handling dynamic environments where manual adjustments would be impractical or too slow."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Of course, results aren\u2019t always guaranteed, but there are plenty of ways to measure and fine-tune models to make them more accurate. It\u2019s also important to take a step back and make sure you clearly understand the problem you\u2019re trying to solve. With the right tools and techniques, ML systems can be adjusted to perform better and deliver more reliable results."})]})})})},{title:"Types of Machine Learning",page:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("h3",{className:"h5",children:"Types of Machine Learning"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Expanding on the previous section, Demystifying AI, machine learning allows systems to learn from data and adapt over time. However, not all learning works in the same way. Machine learning techniques are grouped into three main types, each designed to handle specific kinds of data and tasks: supervised learning, unsupervised learning, and reinforcement learning."}),(0,d.jsxs)("h3",{className:"h5",children:[(0,d.jsx)("img",{src:"images/help-center/supervised.webp",alt:"b",style:{width:"2.5rem",marginRight:"0.5rem"},loading:"lazy"})," Supervised Learning: Learning from Labelled Data "]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"In supervised learning, the model is trained on a dataset where each input comes with a corresponding correct output, known as labelled data. The goal is to learn the relationship between inputs and outputs so the model can make accurate predictions on new, unseen data. It\u2019s often used in tasks like image classification, where the model is trained on images labelled with their correct category, or in email filtering, where messages are labelled as spam or not spam."}),(0,d.jsxs)("h3",{className:"h5",children:[(0,d.jsx)("img",{src:"images/help-center/unsupervised.webp",alt:"b",style:{width:"2.5rem",marginRight:"0.5rem"},loading:"lazy"})," Unsupervised Learning: Finding Patterns in Unlabelled Data "]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Unsupervised learning deals with data that has no labels, meaning the model must find structure and patterns in the dataset without explicit guidance. This type of learning is typically used for clustering or association tasks, where the model groups similar data points together or identifies hidden relationships within the data. A common application is customer segmentation, where businesses group customers based on their purchasing behaviour without having predefined categories."}),(0,d.jsxs)("h3",{className:"h5",children:[(0,d.jsx)("img",{src:"images/help-center/reward.webp",alt:"b",style:{width:"2.5rem",marginRight:"0.5rem"},loading:"lazy"})," Reinforcement Learning: Learning Through Trial and Error "]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Reinforcement learning involves training a model to make decisions by interacting with an environment. The model learns through feedback in the form of rewards or penalties, adjusting its actions to maximise rewards over time. This type of learning is often used in situations where a sequence of decisions needs to be optimised, such as robotics, autonomous driving, or game-playing AI, where the system learns to navigate by trial and error to achieve its goals."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Each type of machine learning offers different advantages depending on the data and the specific task at hand."})]})})})}]},{title:"How AI Works",firstPage:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("img",{src:"images/help-center/howaiworks.webp",alt:"cover",loading:"lazy"}),(0,d.jsx)("h3",{className:"h5"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Artificial Intelligence (AI) operates through the combination of data, algorithms, and computing power to mimic human cognitive processes. The key to AI lies in how it learns and adapts. Instead of following pre-set rules, AI systems analyse vast amounts of data to i
1dentify patterns and make predictions or decisions. These systems rely heavily on machine learning algorithms to detect trends in data and adjust their actions based on those patterns."}),(0,d.jsxs)("h3",{className:"h5",children:[(0,d.jsx)("img",{src:"images/help-center/data.webp",alt:"i",style:{width:"2.5rem",marginRight:"0.5rem"},loading:"lazy"})," Data: The Foundation "]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"AI\u2019s performance is entirely dependent on data. The more data it has, the better it can learn. Data serves as the raw material, feeding AI models information about the world. This data can come from various sources, including images, text, or sensor readings. For example, in natural language processing (NLP), AI systems learn from millions of text documents to understand language structure and meaning."}),(0,d.jsxs)("h3",{className:"h5",children:[(0,d.jsx)("img",{src:"images/help-center/aibrain.webp",alt:"b",style:{width:"2.5rem",marginRight:"0.5rem"},loading:"lazy"})," Algorithms: The Brain of AI "]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Algorithms are the set of instructions that tell the AI system how to process the data. Different types of algorithms are used for different tasks, such as recognising images, translating languages, or predicting future trends. These algorithms work by analysing data, identifying patterns, and using these patterns to make decisions. Machine learning algorithms, particularly, enable AI systems to learn from data and continuously improve over time without needing to be reprogrammed."}),(0,d.jsxs)("h3",{className:"h5",children:[(0,d.jsx)("img",{src:"images/help-center/aimodel.webp",alt:"it",style:{width:"2.5rem",marginRight:"0.5rem"},loading:"lazy"})," Training: The Models "]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"AI systems are trained using datasets, which teach the model how to make accurate predictions or decisions. During the training phase, the AI model processes the data, learns the relationships between input and output, and adjusts its internal parameters to improve its accuracy. Once trained, the model can be applied to new data to make predictions or decisions."}),(0,d.jsxs)("h3",{className:"h5",children:[(0,d.jsx)("img",{src:"images/help-center/aiiterate.webp",alt:"it",style:{width:"2.5rem",marginRight:"0.5rem"},loading:"lazy"})," Continuous Learning and Adaptation "]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"A significant aspect of AI is its ability to learn and adapt over time. As more data is fed into the system, it refines its predictions or decisions. This is particularly crucial in dynamic environments, like coffee Farming where AI systems must adapt to changes without constant human intervention."})]})})}),content:[{title:"The Building Blocks Data, Algorithms, Models",page:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("h3",{className:"h5",children:"The Building Blocks of AI: Data, Algorithms, and Models"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"At the core of any AI system are data, algorithms, and models. These three components work together to enable machines to learn, make decisions, and improve over time."}),(0,d.jsxs)("h3",{className:"h5",children:[(0,d.jsx)("img",{src:"images/help-center/data.webp",alt:"i",style:{width:"2.5rem",marginRight:"0.5rem"},loading:"lazy"})," Data: The Foundation of AI learnng "]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Data is the essential ingredient for AI. It serves as the input that fuels learning and training processes. Whether it\u2019s numbers, text, images, or sensor readings, the data needs to be both extensive and relevant to the task at hand. AI systems rely on large datasets to detect patterns, correlations, and anomalies, and the quality of the data directly influences the accuracy and reliability of the AI\u2019s performance. In the training phase, data is split into training sets and testing sets. The training set is used to teach the algorithm how to recognise patterns, while the testing set evaluates how well the model can generalize to new data."}),(0,d.jsxs)("h3",{className:"h5",children:[(0,d.jsx)("img",{src:"images/help-center/aibrain.webp",alt:"i",style:{width:"2.5rem",marginRight:"0.5rem"},loading:"lazy"})," Algorithms: The Logic Behind Learning "]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Algorithms are the instructions that guide how AI systems learn from data. They are mathematical models designed to process data, learn from it, and optimize the model\u2019s ability to predict or make decisions. Different algorithms are used for different types of learning tasks, from classification to regression to clustering. For instance, algorithms might focus on identifying patterns, predicting future outcomes, or even discovering relationships between variables."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"During training, algorithms adjust their internal parameters to minimize error and improve accuracy. This involves a process of optimization, where the algorithm tunes itself based on the feedback it receives from the data. The training process is iterative, meaning the algorithm repeatedly processes the data, making adjustments until the model\u2019s performance meets a de
1sired threshold."}),(0,d.jsxs)("h3",{className:"h5",children:[(0,d.jsx)("img",{src:"images/help-center/aimodel.webp",alt:"i",style:{width:"2.5rem",marginRight:"0.5rem"},loading:"lazy"})," Training and Models: Building the System "]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"The training process is where the magic happens. During training, the algorithm processes the data and learns how different inputs relate to specific outputs. For example, in a financial prediction system, the algorithm would process historical data to learn which economic factors influence stock prices. The result of this process is a model, which is the system's understanding of how inputs and outputs are connected."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Once the model is trained, it can then be applied to new, unseen data to make predictions or decisions. However, a model\u2019s usefulness doesn\u2019t stop at the training phase. It must be evaluated and fine-tuned over time. This is where the validation phase comes in\u2014evaluating the model on data it hasn\u2019t seen before to ensure that it performs well in real-world situations."}),(0,d.jsxs)("h3",{className:"h5",children:[(0,d.jsx)("img",{src:"images/help-center/aiiterate.webp",alt:"i",style:{width:"2.5rem",marginRight:"0.5rem"},loading:"lazy"})," The Feedback Loop: Learning and Adapting"]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"AI doesn\u2019t just learn once; it continually adapts and improves. When new data is introduced, the model can be retrained or updated, ensuring that its predictions remain accurate even as the environment changes. This continuous learning process is a critical aspect of AI's ability to handle dynamic and evolving tasks. By refining the model through further training and additional data, AI systems become more resilient, responsive, and accurate over time."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Together, data, algorithms, and models form a cyclical system of learning. Data feeds the algorithm, the algorithm processes that data to create a model, and the model is used to make predictions or decisions. Over time, the system learns, improves, and adapts, allowing AI to evolve and remain effective as new challenges and data emerge."})]})})})},{title:"Training AI Models",page:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("h3",{className:"h5",children:"Training AI Models"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Training an AI model is a multi-step process that involves preparing the data, selecting important features, and tuning the model to optimise its performance. Each of these steps is crucial to ensure that the AI system can accurately learn from the data and make reliable predictions."}),(0,d.jsxs)("h3",{className:"h5",children:[(0,d.jsx)("img",{src:"images/help-center/data.webp",alt:"i",style:{width:"1.5rem",marginRight:"0.5rem"},loading:"lazy"})," Data Preparation: Laying the Foundation"]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Before any training can take place, the data must be carefully prepared. Raw data is often incomplete, inconsistent, or contains noise that can mislead the algorithm. The data preparation phase involves several steps to clean, normalize, and format the data so that it is suitable for analysis."}),(0,d.jsx)("h7",{children:"-Cleaning the Data:"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"This step involves handling missing values, removing duplicates, and correcting any errors in the dataset. Inaccurate or incomplete data can negatively impact the model\u2019s performance, so thorough cleaning is essential"}),(0,d.jsx)("h7",{children:"-Normalisation and Scaling Data:"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Different types of data often have different scales, such as age being in years and income being in thousands of dollars. Normalisation ensures that these differences don\u2019t skew the algorithm\u2019s learning by bringing all variables onto a common scale. This step is especially important for algorithms that rely on distance-based calculations, such as many machine learning models."}),(0,d.jsx)("h7",{children:"-Splitting the Data:"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"To evaluate a model\u2019s performance, the data is typically split into training and test sets. The training set is used to teach the model, while the test set is kept aside to validate the model\u2019s performance on unseen data"}),(0,d.jsxs)("h3",{className:"h5",children:[(0,d.jsx)("img",{src:"images/help-center/aibrain.webp",alt:"i",style:{width:"2.5rem",marginRight:"0.5rem"},loading:"lazy"})," Feature Selection: Choosing What Matters"]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Not all data points are equally valuable for making predictions. Feature selection is the process of identifying the most important variables (or \u201cfeatures\u201d) that will contribute to the model\u2019s success. In machine learning, a \u201cfeature\u201d refers to a
1specific variable in the dataset, such as a person\u2019s age, income, or location."}),(0,d.jsx)("h7",{children:"-Reducing Complexity:"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Selecting the right features helps simplify the model by focusing on the variables that matter most. This reduces the complexity of the model, speeds up the training process, and often improves accuracy."}),(0,d.jsx)("h7",{children:"-Avoiding Overfitting: "}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Overfitting occurs when a model becomes too complex and begins to memorise the training data instead of generalizing to new data. By focusing on key features, feature selection helps prevent overfitting, allowing the model to perform well on new data.Feature selection can be done manually, or it can be automated using algorithms that rank features by their importance."}),(0,d.jsxs)("h3",{className:"h5",children:[(0,d.jsx)("img",{src:"images/help-center/aimodel.webp",alt:"i",style:{width:"2.5rem",marginRight:"0.5rem"},loading:"lazy"})," Model Tuning: Fine-Tuning for Optimal Performance"]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Once the data is ready and features are selected, the model is trained using the chosen algorithm. However, training is not a one-size-fits-all process. To achieve the best results, the model needs to be tuned, which involves adjusting the model\u2019s hyperparameters\u2014settings that control the learning process."}),(0,d.jsx)("h7",{children:"-Hyperparameter Tuning: "}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Hyperparameters, such as learning rate or the depth of decision trees, are parameters set before the learning process begins. They are crucial in determining how the model learns from data. For example, a learning rate that is too high might cause the model to miss important patterns, while a rate that is too low might result in slow learning or getting stuck in suboptimal solutions."}),(0,d.jsx)("h7",{children:"-Cross-Validation: "}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"This is a technique used to evaluate model performance by splitting the training data into multiple subsets and training the model on different combinations of these subsets. This ensures that the model isn\u2019t just performing well on a specific portion of the data but can generalize across the entire dataset."}),(0,d.jsx)("h7",{children:"-Iterative Training: "}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Model tuning is an iterative process. The model is trained, evaluated, and then retrained with adjustments made to hyperparameters or features. This cycle continues until the model reaches the desired level of accuracy and robustness."}),(0,d.jsxs)("h3",{className:"h5",children:[(0,d.jsx)("img",{src:"images/help-center/aiiterate.webp",alt:"i",style:{width:"2.5rem",marginRight:"0.5rem"},loading:"lazy"})," Evaluating the Model:"]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Testing for Real-World Performance Once the model is trained and tuned, it\u2019s tested on the unseen data from the test set. This step is critical to ensure that the model can generalise to new situations. If the model performs well on the test set, it indicates that the model has learned effectively from the training data without overfitting."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Training an AI model involves cleaning and preparing the data, selecting the most relevant features, and fine-tuning the model to optimise performance. It\u2019s a process of continuous refinement, ensuring the AI can make accurate and reliable predictions when deployed in real-world scenarios."})]})})})}]},{title:" Evaluating AI Performance",firstPage:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("h3",{className:"h5",children:"Evaluating AI Performance"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Evaluating an AI model's performance is a critical step in ensuring that it can accurately make predictions and generalize to new, unseen data. The goal is not just to achieve good results on training data but to ensure the model works effectively in real-world scenarios."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"The evaluation process typically begins by splitting the data into training and test sets. The model is trained on one set of data and evaluated on a separate set to see how well it performs on new input
1s. Various performance metrics are used to measure the model\u2019s success, such as accuracy, precision, and recall, depending on the task at hand."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"It\u2019s also essential to check for common issues like overfitting, where the model performs well on training data but poorly on new data, or underfitting, where the model fails to capture important patterns. Evaluating AI performance helps identify these issues early and ensures that the model is robust and ready for real-world application"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Beyond accuracy, AI systems must be tested for their ability to generalize\u2014to handle variations in data and changing environments. Continuous monitoring and evaluation are vital even after deployment, ensuring that models remain effective as they encounter new or evolving data over time."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"In the following sections, we\u2019ll explore specific metrics and methods to measure and fine-tune AI model performance in greater detail."})]})})}),content:[{title:"Measuring Accuracy",page:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("h3",{className:"h5",children:"Measuring Accuracy"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Measuring the success of an AI model is key to understanding how well it performs and whether it can be trusted in real-world applications. Accuracy is one of the most commonly used metrics, representing the percentage of correct predictions made by the model out of all predictions. However, accuracy alone doesn\u2019t always provide the full picture, especially when dealing with imbalanced datasets or specific problem domains where false positives or false negatives carry significant weight."}),(0,d.jsx)("h3",{className:"h5",children:"Recall: "}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Also known as sensitivity, recall measures how well the model identifies all actual positive instances. It answers the question: of all the true positives in the data, how many did the model successfully catch? High recall is important in cases where missing a positive case (like failing to detect fraud) could be highly problematic."}),(0,d.jsx)("img",{src:"images/help-center/recall.webp",alt:"cover",loading:"lazy"}),(0,d.jsx)("h3",{className:"h5",children:"F1 Score: "}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"The F1 score is a balance between precision and recall, providing a single metric to evaluate models that may need to manage trade-offs between the two. It\u2019s particularly useful when you need to balance precision and recall rather than optimising one at the expense of the other."}),(0,d.jsx)("img",{src:"images/help-center/fscore.webp",alt:"cover",loading:"lazy"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"These metrics together offer a more complete understanding of an AI model\u2019s performance. While accuracy provides a general sense of correctness, precision and recall dive deeper into how well the model handles different types of errors, allowing for more nuanced evaluation and fine-tuning of the system."})]})})})},{title:"Continuous Model Improvement",page:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("h3",{className:"h5",children:"Why Continuous Improvement Matters"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"AI models don\u2019t exist in a vacuum\u2014they need to keep up with the real world. Over time, customer behaviors change, new trends emerge, and the data the model was trained on might no longer reflect current realities. If the model isn\u2019t updated regularly, its predictions can become less accurate."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"For instance, think about climate and weather. The effects on coffee farm health and related yield estimates can vary significantly depending on seasonal impacts, microclimates, and altitudes. Models that fail to account for these changing conditions risk producing irrelevant or inaccurate predictions, leading to missed opportunities for farmers and businesses."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Or take a fraud detection system in banking. Fraudsters constantly change their tactics to bypass security measures. If the system relies on outdated data, it won\u2019t catch new types of fraud, leaving businesses and customers vulnerable."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"The chart below shows how a model\u2019s performance can decline over time as its environment changes. You can see how retraining events\u2014when the model is updated with new data\u2014help restore its effectiveness and keep it on track."}),(0,d.jsx)("img",{src:"images/help-center/mp.webp",alt:"cover",loading:"lazy"}),(0,d.jsx)("h3",{className:"h5",children:"Why Continuous Improvement is Essential "}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Continuous improvement is essential to prevent this kind of performance decline. By regularly updating models to reflect the latest data and trends, we can ensure they stay accurate, relevant, and effective in the face of change."})]})})})}]},{title:"Retraining Models with New Data",firstPage:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("h3",{className:"h5",children:"Retraining Models with New Data"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"One of the key methods to continuously improve an AI model is retraining. As new data becomes available, the model is re-ex
1posed to the updated information, allowing it to adapt to recent trends, behaviours, or patterns that were absent in the original dataset. Retraining helps the model stay aligned with the latest data and ensures that it can handle emerging challenges and opportunities."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"This process typically involves feeding new data back into the model, adjusting its parameters, and refining its performance. Depending on the application, retraining can be performed periodically (e.g., monthly or quarterly) or in real time, where models are continuously updated as new data streams in."})]})})}),content:[{title:"Fine-Tuning and Hyperparameter Optimisation",page:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("h3",{className:"h5",children:"Fine-Tuning and Hyperparameter Optimisation"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Beyond retraining, fine-tuning the model's hyperparameters can lead to performance improvements. Hyperparameters\u2014such as the learning rate, batch size, or model complexity\u2014control how the model learns from the data. Adjusting these settings, based on feedback from the model\u2019s performance, allows for more targeted improvements and can help avoid issues like overfitting or underfitting."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"For example, a model trained to predict sales trends may perform well initially, but over time, as seasonal patterns shift, fine-tuning its parameters can help the model adapt to these new variations, maintaining its accuracy."}),(0,d.jsx)("img",{src:"images/help-center/epochchart.webp",alt:"cover",loading:"lazy"})]})})})},{title:"Monitoring and Feedback Loops",page:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("h3",{className:"h5",children:"Monitoring and Feedback Loops "}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Continuous improvement also requires monitoring the model\u2019s performance in real time, even after deployment. Monitoring helps track how well the model is performing in production environments, identifying areas where its predictions are falling short or where new patterns have emerged. Feedback loops can be created, where the model\u2019s decisions are assessed, and corrections or updates are made accordingly."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"This active learning process allows the AI system to keep pace with real-world changes, improving its ability to generalize and handle new types of data effectively. Monitoring also ensures that the model remains compliant with performance goals, ethical standards, and fairness considerations."}),(0,d.jsx)("img",{src:"images/help-center/feedbackloop.webp",alt:"cover",loading:"lazy"})]})})})},{title:"Avoiding Model Degradation",page:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("h3",{className:"h5",children:"Avoiding Model Degradation"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Without continuous updates, even the best AI models will experience model degradation, where the quality of their predictions decreases as the data changes. Regular updates and retraining cycles help prevent this, ensuring that the model remains useful, accurate, and capable of delivering reliable results in a constantly shifting environment."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"In essence, continuous model improvement is not just a best practice but a necessity for any AI system that operates in dynamic, real-world conditions. By retraining with new data, fine-tuning model parameters, and closely monitoring performance, AI systems can evolve and maintain high levels of accuracy and effectiveness over time."}),(0,d.jsx)("img",{src:"images/help-center/degredation.webp",alt:"cover",loading:"lazy"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"In the following sections, we\u2019ll dive into the specific applications of AI in agriculture, exploring how continuous learning and model improvement can help optimise farming practices. This ongoing refinement is key to developing AI systems that enhance productivity, sustainability, and decision-making in modern agriculture."})]})})})}]},{title:"AI Applications in Coffee",firstPage:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("img",{src:"images/help-center/aiapplicationincoffee.webp",alt:"cover",loading:"lazy"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"AI has proven highly effective in large-scale, controlled agricultural environments where automation and precision farming are easier to implement. However, in more complex supply chains like coffee, where smallholder farmers operate fragmented plots in difficult terrains and high altitudes, AI faces unique challenges."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"In coffee farming, climatic variability adds another layer of complexity. The decentralised nature of coffee production, coupled with rapidly changing environmental conditions, makes it difficult for standard AI models to manage inconsistent and incomplete data. These fragmented supply chains and variable climates require additional tools to gather, clean, and model data effectively."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"By integrating 13 years of experience, we have developed the tools to make AI work in these complex environments. By combining AI with advanced data management and predictive tools, we can address the challenges of fragmented supply chains and climatic unpredictability. This allows for better resource management, improved decision-making, and more resilient supply chains, even in the most difficult environments."})]})})}),content:[{title:"Coffee Yield Estimates",page:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("h3",{className:"h5",children:"Coffee Yield In Season Estimates"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Estimating coffee yields is a complex process influenced by variables like tree health, weather patterns, farm practices, and soil health. Manual in-field methods exist, such as flowering and branch counts, but these are time-consuming in large areas and difficult geographies. As such, they are rarely attempted by washing stations due to their impracticality."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Using AI and machine learning (ML), we have developed a system that combines historical trends, real-time data, and predictive modelling. This approach not only improves accuracy but also allows for ongoing refinement with feedback loops, ensuring the system adapts to evolving conditions like climate variability and changing farming practices. We have monitored thous
1ands of farms over 10 years with key sets of data for training, including results from our demo farms and their yields."}),(0,d.jsx)("img",{src:"images/help-center/cwsstats.webp",alt:"cover",loading:"lazy"}),(0,d.jsx)("h3",{className:"h5",children:"Mapping Farms & getting regular Clean data"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Thirteen years ago, we took a first-principles approach to identify the types of data required to train models for estimating yields\u2014both at the farm level and across entire washing station zones. This began with digitally mapping farms using GPS waypoints to record boundaries, physically count trees, assess tree ages, and capture farmer details. This geolocation-based mapping created a foundational asset where all subsequent data could be tied directly to individual farms."}),(0,d.jsx)("h3",{className:"h5",children:"Regular Farm Visits and Scouting new data in Farms"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Regular farm visits and scouting efforts recorded incidences of pests and diseases, including when and where they occurred. These were tagged to specific farms, building a detailed timeline of challenges faced by each plot. Our custom WRF model provided key weather variables, while satellite imagery targeted over these farms supplied data on biomass health and stress levels. Together, these inputs created a comprehensive view of each farm\u2019s conditions."}),(0,d.jsx)("img",{src:"images/help-center/infields.webp",alt:"cover",loading:"lazy"}),(0,d.jsx)("h3",{className:"h5",children:"Collecting Coffee Cherry Yield over farms"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Cherry yield records from these farms, delivered to washing stations, formed a key performance metric\u2014a kind of \u201creward\u201d or health score within an unsupervised learning framework. Satellite-derived metrics like NDVI provided additional insights into farm health and stress. Over time, we found that time itself was a critical variable, helping to identify cycles and inconsistencies, such as recovery periods inherent to coffee trees and stresses caused by drought, pests, and diseases during specific seasons."}),(0,d.jsx)("h3",{className:"h5",children:" Demo Farms as Benchmarks"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Our demo farms, distributed across diverse geographies and altitudes, provide benchmarks for what\u2019s achievable under optimal conditions. These farms implement GAP77 practices, significantly outperforming smallholder farms. For instance:"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:(0,d.jsx)("strong",{children:"1. Smallholder farms typically produce 0.3\u20130.5 kg of cherry per tree annually. before GAP77"})}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:(0,d.jsx)("strong",{children:"2. Demo farms achieve 6kg of cherry per tree in the same conditions."})}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:(0,d.jsx)("strong",{children:"3. Newly planted demo farms achieve 2kg year 3, 5kg+ year 4, and 6kg+ subsiquent years again in the same conditions."})}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"By monitoring demo farms from planting to harvest, we grew to establish a baseline for the models given the same conditions and provide smallholders with tangible examples of potential improvements."}),(0,d.jsx)("h3",{className:"h5",children:" Visualising the yield Estimates"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:'The "yield wheel" is a core tool for communicating predictions to farmers CWS and others. It breaks down yields into three key categories:'}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"Potential Yield:"})," What could be achieved with full GAP77 implementation"]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"Potential Loss: "}),"Predicted yield based on current conditions."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"Expected Yield:"}),"Predicted yield based on current conditions & health of farm."]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"By monitoring demo farms from planting to harvest, we grew to establish a baseline for the models given the same conditions and provide smallholders with tangible examples of potential improvements."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Farmers particulary respond well too the wheel on their farm for its clarity and relevance, agreeing with the estimate and often expressing a desire to understand how to close the gap between expected and potential yields. The visualisation also helps align expectations with achievable outcomes."}),(0,d.jsx)("h3",{className:"h5",children:" Challenges and Refinements"}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:["Coffee Yield estimation is a dynamic process that requires constant refinement to address the complexities of coffee farming.",(0,d.jsx)("strong",{children:"Key challenges include:"})]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"Side-Selling: "})," Farmers sometimes sell their cherry to other washing stations, resulting in incomplete yield data and complicating accuracy."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"Seasonality: "}),"The cyclical nature of coffee tree production, with main and recovery seasons, creates natural fluctuations in yields that must be accounted for."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"Climate Variability: "}),"Unpredictable weather patterns, including prolonged droughts, heavier rains, and shifting seasons, introduce uncertainties that demanded adaptive modelling."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"Pests and Diseases "}),"Stress from issues like leaf rust not only reduces immediate yield but also has compounding long-term impacts on tree health and productivity."]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Our models currently achieve approximately 80% accuracy, but this is an evolving metric. As new data is collected and incorporated\u2014particularly in response to climate variability and farm-level changes\u2014our system continues to refine and adapt, ensuring it remains reliable and responsive to real-world challenges."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"These and many other considerations must be taken into account, but our processes are designed to enable scaling across other coffee-producing regions,"})]})})})},{title:"Pest and Disease Alerts",page:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("h3",{className:"h5",children:"Pest and Disease Alerts: Proactive Monitoring and Action"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Coffee farming is highly susceptible to pests and diseases, which can devastate yields and quality if not addressed promptly. Key threats include leaf rust, antestia bug, and wilt, all of which require timely intervention to mitigate their impacts. Our AI-driven pest and disease alert system monitors farms in real-time, using satellite data, weather models, and field reports to predict and respond to these risks effectively"}),(0,d.jsx)("h3",{className:"h5",children:"Event Templates and Monitoring"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"At the heart of our alert system are event templates, which model the conditions that lead to pest or disease outbreaks. These templates are based on years of data collection and focus on key variables, including:"}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:" \u2022 Temperature ranges"})," and leaf wetness duration (e.g., ideal conditions for fungal diseases like leaf rust)."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:" \u2022 Degree days"})," that influence pest development cycles, such as the emergence of antestia bugs."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:" \u2022 Satellite stress indicators"})," like NDVI, to assess overall farm health."]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Our system monitors these variables in real-time across all farms, comparing them to historical conditions where incidents occurred."}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"\u2022 Leaf Rust: "})," When models detect conditions conducive to leaf rust, such as high humidity and extended leaf wetness, farmers receive alerts recommending preventive measures like application of ProBald or targeted treatments for management."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"\u2022Antestia Bug:"})," Alerts are triggered during periods favourable to pest emergence, advising farmers to scout for eggs under leaves and destroy them before they hatch."]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"These insights were validated through scouting visits, where our teams record the presence and severity of incidents in the field via a mobile app. The feedback loop from these reports continuously improves the accuracy of the alerts.The alerts & associated practises in dealing with alerts were backup by regular monthly traning so farmer understand better the SMS alerts- important so we can better understand the impact of the prctises."}),(0,d.jsx)("img",{src:"images/help-center/mapping1.webp",alt:"cover",loading:"lazy"}),(0,d.jsx)("h3",{className:"h5",children:"Feedback Loops for Continuous Improvement"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"The system\u2019s strength lies in its feedback loops. Each pest or disease incident recorded during a scouting visit is timestamped and tagged with variables like severity, location, and environmental conditions. This allows us to:"}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"\u2022 Look back "})," at all teh potentialthe variables that likely caused the incident."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"\u2022Refine event templates"})," to improve future alerts"]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"For instance, if an outbreak of leaf rust occurs, the system analyses historical temperature, humidity, and satellite stress data for the affected farm, identifying patterns that might indicate future risk. Over time, this iterative learning process enhances the system\u2019s precision and scalability."}),(0,d.jsx)("h3",{className:"h5",children:"Validating Alerts"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Validation is a critical step in maintaining the reliability of the system. For years, we have scouted farms and demo plots, using the app to record:"}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"\u2022 Confirmation of alerts: "})," Whether the predicted risk materialised on the farm."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"\u2022Effectiveness of recommended actions:"})," Whether the tasks tied to alerts reduced the impact of the outbreak."]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"This rigorous process partnering with farmers ensures that alerts are not only timely but also actionable and accurate, building trust among farmers."}),(0,d.jsx)("h3",{className:"h5",children:"Proactive Management of Threats to Yield"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Our pest and disease alert system enables farmers to actively manage risks before they escalate into significant yield losses. By combining real-time monitoring, historical data, and practical, actionable solutions, the system bridges the gap between prediction and prevention."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"For coffee companies, this approach plays a crucial role in aligning with ESG (Environmental, Social, and Governance) standards by addressing key sustainability goals. By Proactively managing pest and disease threats supports environmental stewardship by minimising chemical interventions and preserving biodiversity. Socially, it empowers farmers with knowledge and tools, fostering resilience and improving livelihoods. Governance is strengthened through traceable, data-driven insights that ensure transparency and accountability across the supply chain."})]})})})},{title:"Coffee Tree detection",page:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("h3",{className:"h5",children:" Coffee Tree Detection: Laying the Foundation for Data-Driven Coffee Production "}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Coffee tree detection is a foundational step in understanding the state of national coffee production and critical application of AI and machine learning (ML) in the agricultural sector, provides unprecedented accuracy and efficie
1ncy in tracking coffee production. Accurate detection enables governments, stakeholders, and coffee companies to make informed decisions about resource allocation, productivity improvements, and sustainability goals."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Traditional methods, such as drive-by surveys and manual counting, are costly, time-consuming, and often unreliable. By combining AI, machine learning (ML), and satellite imagery, CONCENSUS-AI\u2122 offers a precise, efficient, and scalable alternative, providing actionable insights into coffee tree stock, health, and productivity."}),(0,d.jsx)("h3",{className:"h5",children:" Where it started "}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Our exploration into AI-driven coffee tree detection began in 2015, when we partnered with the Burundi government and the World Bank to trial an early coffee tree detection algorithm."}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Today, leveraging vast data sets and advancements in AI and machine learning, our models provide highly accurate assessments. Drawing on years of data from mapped coffee farms across multiple geographies, these models estimate tree stock, age, and health, delivering a comprehensive snapshot of the country\u2019s coffee production."}),(0,d.jsx)("h3",{className:"h5",children:"How AI and ML Drive Coffee Tree Detection "}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"The backbone of our coffee tree detection system lies in advanced AI and ML algorithms that process vast amounts of geospatial data. These technologies enable accurate and scalable tree detection by leveraging:"}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"\u2022 Image Recognition and Classification: "})," Our ML models are trained to analyse high-resolution satellite imagery, identifying coffee trees based on features like canopy structure, spectral signatures, and colour intensity. This allows for precise detection even in intercropped or dense vegetation."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"\u2022 Temporal Modelling: "})," Coffee trees exhibit cyclical growth patterns tied to A&B seasons. By incorporating time as a variable, our models predict tree productivity and health, accounting for seasonal variations and growth cycles."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"\u2022 Multispectral Analysis: "})," Using NDVI (Normalised Difference Vegetation Index) and hyperspectral imagery where needed, the system assesses vegetation health and detects stress indicators These insights are tied directly to tree-level detection for a complete farm health overview."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"\u2022 Feedback Loops: "})," Validating satellite data with on-the-ground observations ensures continuous improvement. Feedback from farm scouting visits helps refine detection models, ensuring they stay aligned with real-world conditions"]}),(0,d.jsx)("h3",{className:"h5",children:"Types and Volume of Data"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"The backbone of our coffee tree detection system is the extensive satellite imagery and data collected over the past eight years across 250,000 coffee farms in four countries. This data, combined with advanced AI and ML algorithms, enables us to deliver highly accurate tree detection at scale."}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"\u2022 Satellite Imagery: "})," Our models are trained to utilise all available satellite data, processing features like canopy structure and vegetation patterns to detect coffee trees. The data includes regular high-resolution imagery, spanning years of continuous monitoring."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"\u2022 Temporal Coverage: "})," With eight years of consistent satellite imagery over mapped coffee farms, the models are trained on regular updates, allowing for a comprehensive understanding of tree patterns and distributions."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"\u2022 Validation: "})," As part of the process, ground validation is conducted on 10% of detected areas in each region, ensuring accuracy through direct feedback loops. Where gaps exist, such as newly detected farms, they are mapped using a mobile app to enhance future datasets."]}),(0,d.jsx)("h3",{className:"h5",children:"Challenges in Coffee Tree Detection"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Coffee tree detection faces unique challenges that require continual refinement"}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"\u2022 Unproductive trees: "})," Some coffee trees have so little foliage\u2014often just sticks in the ground\u2014that they go undetected and yield almost nothing. These typically result from poor farm management, and since they don\u2019t contribute to production, they shouldn\u2019t be counted as part of the productive stock."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"\u2022 Very Young Trees: "})," those under 1 years old will not be detected due to their size. Higher resolution satellite data can address this, though it comes with increased costs at scale. these trees are also tend not to be productive at this age."]}),(0,d.jsxs)("p",{className:"fs-7 text-dark",children:[(0,d.jsx)("strong",{children:"\u2022 Validation: "})," We have established this by leveraging feedback loops, robust datasets, and adaptable algorithms, our system addresses these challenges to provide reliable, scalable tree detection globally."]}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"By combining advanced AI and ML algorithms with extensive datasets and validation processes, coffee tree detection systems deliver reliable, scalable, and actionable insights. While challenges such as unproductive and
1young trees exist, the continuous refinement of models ensures adaptability to diverse geographies and farming conditions. This approach not only supports accurate tree numbers and tree ages but also lays the groundwork for a sustainable, well-managed coffee sector."})]})})})}]},{title:"AI Transparency and Bias",firstPage:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("h3",{className:"h5",children:"AI Transparency and Bias Mitigation"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"AI transparency and bias mitigation is a crucial aspect of modern AI systems. It\u2019s not just \u201cmanagement speak\u201d\u2014it tackles real-world challenges and responsibilities, particularly in contexts that affect people\u2019s lives. Ensuring transparency and fairness helps build trust and accountability, leading to AI solutions that are both equitable and effective."}),(0,d.jsx)("h3",{className:"h5",children:"Why Transparency and Bias Mitigation Matter"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Bias can occur when AI relies on unbalanced or incomplete data, potentially harming certain groups or unfairly favouring others. Transparency addresses this by making the decision-making process understandable to users, stakeholders, and impacted communities. Together, these principles ensure AI systems are fair, trustworthy, and able to deliver meaningful real-world outcomes."}),(0,d.jsx)("h3",{className:"h5",children:"Getting Farmers Fair, Transparent insights"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"By emphasising data integrity, openness, and real-world engagement, we can have smallholder farmers benefit equally from modern technology. Our aim is to create solutions that are both inclusive and impactful, ensuring no one is left behind in the push towards smarter, data-driven agriculture."})]})})}),content:[{title:"We Take a Principle-First Approach",page:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("h3",{className:"h5",children:" We Take a Principle-First Approach"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"We prioritise transparency and fairness in every phase of development, recognising that these values are essential for making unbiased, equitable decisions. Through rigorous data collection and regular audits of our datasets, we reduce the risk of biased recommendations that might otherwise disproportionately affect certain farmers or regions. This principle-first mindset shapes our entire AI life cycle\u2014from how we source data to how we evaluate model performance."}),(0,d.jsx)("h3",{className:"h5",children:" On-the-Ground Involvement for Real Insights "}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"We believe fairness begins with understanding. That\u2019s why we work directly with farmers, immersing ourselves in their day-to-day challenges. This firsthand perspective helps us uncover potential biases in our models and refine them for truly equitable, effective solutions. By being present in the field, we ensure that our tools address real needs rather than theoretical assumptions."})]})})})},{title:"Continuous Learning Feedback Loops",page:(0,d.jsx)(r.A,{lg:"9",children:(0,d.jsx)("div",{className:"nl-entry-wrap pt-3 pslg-5",children:(0,d.jsxs)("div",{className:"nl-entry",children:[(0,d.jsx)("h3",{className:"h5",children:" Continuous Learning Through Feedback Loops"}),(0,d.jsx)("p",{className:"fs-7 text-dark",children:"Addressing bias and refining fairness isn\u2019t a one-time task\u2014it\u2019s an ongoing process. We rely on continuous feedback loops, val
1idating our models in actual farming conditions. When a recommendation doesn\u2019t meet expectations, we incorporate those results back into our system to improve future performance. This iterative approach keeps our AI relevant and fair across diverse geographies, environmental conditions, and farming practices."})]})})})}]}];return(0,d.jsxs)(n.A,{children:[(0,d.jsx)(r.A,{lg:"3",children:(0,d.jsx)("div",{className:"nl-entry-sidebar",children:(0,d.jsx)("ul",{className:"nl-list-link nl-list-link-collapse flush",children:p.map(((a,s)=>(0,d.jsxs)("li",{className:"align-items-start",children:[(0,d.jsxs)(i.N_,{to:"#",className:"nl-list-link-collapse-link w-100 text-start gap-2 p-0 border-0 bg-transparent",onClick:()=>h(s),"aria-expanded":e===s,"aria-controls":`accordion-${s}`,children:[(0,d.jsx)("span",{className:"fs-7",children:a.title}),(0,d.jsx)("em",{className:"icon ni "+(e===s?"ni-minus":"ni-plus")})]}),(0,d.jsx)("div",{id:`accordion-${s}`,className:"collapse "+(e===s?"show":""),"aria-labelledby":`accordion-${s}-heading`,children:(0,d.jsx)("ul",{className:"nl-list-dot nl-list-dot-sm pt-2",children:a.content.map(((e,a)=>(0,d.jsx)("li",{className:"nl-list-dot-item align-items-start",children:(0,d.jsx)(i.N_,{onClick:()=>m(a),to:"#",children:e.title})},a)))})})]},s)))})})}),-1===o?p[s].firstPage:p[s].content[o].page]})}const h=function(){return(0,d.jsxs)(l.A,{title:"What is AI and ML",description:"An explainer on how artificial intelligence and machine learning are applied to coffee agriculture \u2014 yield prediction, pest modelling, and satellite health analysis.",children:[(0,d.jsxs)(o.Cu,{className:"pt-120 pt-lg-180",masks:["blur-1 left top","blur-1 right bottom"],children:[(0,d.jsxs)("div",{className:"nl-section-head",children:[(0,d.jsx)("nav",{children:(0,d.jsxs)("ol",{className:"breadcrumb mb-3 mb-md-4",children:[(0,d.jsx)("li",{className:"breadcrumb-item",children:(0,d.jsx)(i.N_,{to:"/blog",children:"Blog"})}),(0,d.jsx)("li",{className:"breadcrumb-item active","aria-current":"page",children:"What is AI & ML"})]})}),(0,d.jsx)("h1",{children:" What is AI & ML?"}),(0,d.jsx)("hr",{className:"mt-5 mb-0"})]}),(0,d.jsx)(o.Cu.Content,{children:(0,d.jsx)(c,{})})]}),(0,d.jsx)(o.Cu,{children:(0,d.jsx)(o.Cu.Content,{children:(0,d.jsx)("div",{className:"bg-primary-gradient rounded-3 p-5 p-lg-7",children:(0,d.jsxs)(n.A,{className:"align-items-center gy-5",children:[(0,d.jsx)(r.A,{lg:"8",children:(0,d.jsxs)("div",{className:"d-flex flex-column gap-3 flex-lg-row align-items-center",children:[(0,d.jsx)(o.VN,{size:"xl",rounded:"circle",variant:"white",icon:"chat-fill text-primary"}),(0,d.jsxs)("div",{className:"text-center text-lg-start",children:[(0,d.jsx)("h3",{className:"text-white text-capitalize m-0",children:"Chat with our support team!"}),(0,d.jsx)("p",{className:"fs-7 mb-0 text-white opacity-75",children:"Get in touch with our support team if you still can\u2019t find your answer."})]})]})}),(0,d.jsx)(r.A,{lg:"4",className:"text-center text-lg-end",children:(0,d.jsx)(o.Hl,{href:"/contact-us-solution",className:"btn-white text-dark fw-semiBold",label:"Contact Support"})})]})})})}),(0,d.jsx)(o.Cu,{className:"pb-lg-0",children:(0,d.jsxs)(n.A,{className:"justify-content-center justify-content-lg-between align-items-center gx-0 pb-5 border-bottom border-lighter",children:[(0,d.jsx)(r.A,{lg:"6",xl:"4",children:(0,d.jsxs)("div",{className:"text-center text-lg-start pb-5 pb-lg-0",children:[(0,d.jsx)("h4",{className:"text-capitalize",children:"Subscribe to our newsletter"}),(0,d.jsx)("p",{className:"fs-7",children:"Join the people wanting the latest in Coffee Crop Research"})]})}),(0,d.jsx)(r.A,{md:"10",lg:"6",xl:"5",children:(0,d.jsx)(o.rj,{})})]})})]})}}}]); 2//# sourceMappingURL=3624.4185bc35.chunk.js.map
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.