7.4. Components of AI Modelling

The foundation of every model. It includes: • Input features (independent variables) • Labels/output (dependent variable in supervised learning)

Data is the starting point for creating any AI model. It consists of two main parts: input features and labels. Input features are the variables the model uses to learn—these are the attributes that will help the model understand the data. For example, if we're modeling fruit identification, the color, weight, and shape of the fruit would be input features. Labels, on the other hand, are the outcomes we want to predict, such as whether a fruit is an apple or an orange in our example of supervised learning.

Think of data as ingredients in a recipe. Just as a good recipe requires the right ingredients (like flour, sugar, and eggs) to make a delicious cake, a good AI model needs the right data to make accurate predictions. If you have spoiled ingredients, the cake will not turn out well, just like a model that uses poor quality data won’t perform effectively.

The mathematical method or formula used to train the model. Examples: • Linear Regression • Decision Trees • K-Nearest Neighbours (KNN) • Support Vector Machines (SVM)

An algorithm in AI acts like a set of cooking instructions that tells the model how to process the input data to learn patterns. Each algorithm has its own strengths and is suited for different types of problems. For instance, Linear Regression is great for continuous data, while Decision Trees can handle categorical data effectively. When you apply these algorithms to the data, you are essentially training your model to perform specific tasks, whether it be predicting prices or classifying items.

Consider algorithms like different styles of cooking. For example, baking uses specific measurements and times (like Linear Regression), while grilling (like Decision Trees) allows for a bit more improvisation based on the ingredients you have. Depending on the desired result (the model's task), you’d choose one style over another.

The outcome of applying an algorithm on data. It is now capable of: • Recognizing patterns • Making predictions or classifications

After training an AI model with data and applying an algorithm, the model is essentially a refined tool that can identify patterns within the data. This means it doesn't just memorize inputs but understands how to predict or classify new data it hasn't seen before. For instance, after training with various fruit data, our model can now easily identify whether a new fruit is an apple or an orange, based on learned characteristics.

Imagine a student learning about different animals in a biology class. Once they study various characteristics—like fur patterns, size, and habitat—they become able to identify animals they've never seen based on the knowledge they've gained. Similarly, an AI model learns from its training data and can make predictions about new, unseen data.

• Training: Feeding the model with known data to learn. • Testing: Checking model’s performance on unseen data.

Training a model involves supplying it with known data so it can learn the relationships between input features and predictions. This is where the model forms its understanding. Testing assesses how well that model can generalize its learning to new data it hasn’t encountered yet. A good model performs well not just on training data but also on unseen data, indicating it has learned properly.

Think of training and testing like rehearsing for a performance. During practice (training), actors learn their lines and cues (model learns the data). The actual performance (testing) demonstrates how well they've remembered and can use that knowledge when faced with a live audience (new data). If they falter during a performance, it indicates areas where they may need more practice.

--