7.4. Modelling

Modelling is the process of training an AI algorithm using the acquired and cleaned data to predict or classify future data points.

Modelling is a crucial step in the AI project cycle where we take the data we've collected and cleaned to create a model that can make predictions. Think of it as teaching a machine to recognize patterns in the data, which allows it to understand and predict future outcomes based on the data it has worked with.

Imagine teaching a child to recognize different fruits. At first, you show them pictures of apples, bananas, and oranges, explaining the differences. After enough examples, the child learns to identify fruits on their own, even new ones they haven't seen before, much like a model learns from data.

Types of AI Models:

1. Supervised Learning – Labeled data used for prediction/classification
2. Unsupervised Learning – Patterns discovered from unlabeled data
3. Reinforcement Learning – Learning through rewards and penalties

There are three main types of AI models. In supervised learning, the model learns from labeled data, where it knows what the output should be. In unsupervised learning, the model analyzes data without predefined labels, looking for patterns on its own. Reinforcement learning involves training the model through rewards and penalties to learn the best actions to take in a given situation.

Think of supervised learning like a teacher helping a student with homework by providing answers. Unsupervised learning is like providing a puzzle without a reference image and letting the person figure it out. Reinforcement learning is similar to training a pet with treats; they learn which actions lead to receiving treats (positive reinforcement) or avoiding negative consequences (penalties).

Steps:

1. Splitting Data – Training and Testing sets
2. Choosing the Algorithm – Decision Trees, SVM, KNN, etc.
3. Training the Model
4. Evaluating the Model – Accuracy, Precision, Recall, F1 Score

The modeling process consists of several key steps. First, we split the dataset into two parts: one for training the model and one for testing its performance. Next, we choose an algorithm suitable for our problem, such as Decision Trees or Support Vector Machines (SVM). After this, we train the model on the training data. Finally, we evaluate the model using metrics like accuracy, precision, recall, and F1 score, which help us understand how well our model performs.

Consider building a model like a two-part exam. The first part is practice, where you learn the material (training), and the second part is a test where you show what you have learned (testing). You would need to pick the right study methods (algorithm) and measure your performance afterwards to see how well you grasped the material.

Important Concepts: • Overfitting and Underfitting • Cross-validation • Bias-Variance Tradeoff

Understanding key concepts in modeling is vital. Overfitting happens when a model learns too much detail from the training data, making it perform poorly on new data. Underfitting occurs when a model is too simple to capture the underlying trends in the data. Cross-validation is a technique to improve model accuracy by splitting the data into different sets for testing to ensure the model's performance is reliable. The bias-variance tradeoff is about finding the right balance between making accurate predictions and ensuring the model is generalizable to new data.

Imagine a tailor making clothes. If they make the clothes too fitted (overfitting), they won't fit well on a different body shape. If the clothes are too loose and basic (underfitting), they won’t look good on anyone. Cross-validation is like getting feedback from several customers before finalizing the design, ensuring it meets a variety of needs. The bias-variance tradeoff is akin to balancing style and comfort in clothing — you want the clothes to look good universally, not just on one person.

--