Fundamentals of Machine Learning Data

1. The Building Blocks of Data: Features and Labels

In supervised machine learning, we start with a dataset. Think of this dataset as a table or a spreadsheet. The columns in this table represent the properties of what we're studying, and we divide them into two categories: features and labels.

• Features: These are the input variables or the characteristics of our data that we use to make a prediction. They are the "clues" we give the machine. If we are trying to predict the price of a house, the features might include its size in square feet, the number of bedrooms, and its age.

• Label (or Target): This is the output variable, the value we are trying to predict. It is the "answer" that the machine learns to produce. For the house price example, the label is the actual selling price of the house.

Essentially, the machine learning process learns the relationship between the features and the label.

The nature of the label determines the type of problem, connecting back to our previous lesson:

• If the label is a continuous number (like price or temperature), it's a regression problem.
• If the label is a distinct category (like 'spam'/'not spam' or 'cat'/'dog'), it's a classification problem.

Features and labels can also be of different types (continuous, categorical, etc.). To explore these distinctions in more detail, please watch the following video.

2. The Learning Process: Training Data and Models

Now that we understand features and labels, let's see how they are used.

• Training Data: This is the dataset used to "teach" the machine learning algorithm. It is a collection of examples where we have both the features and their corresponding correct labels. The algorithm iterates through this training data to find patterns and learn the mapping between the features and the label. Think of it as a student's textbook filled with practice problems and their solutions.

• Model: The model is the output of the training process. It is a mathematical representation of the patterns and relationships the algorithm has learned from the training data. You can think of it as the "knowledge" the system has acquired. Once trained, this model can take new, unseen features as input and predict a label.

For example, after training on house price data, a simple linear regression model might be the equation:

In this equation, the numbers (150, 50000, -1000, 25000) are the learned parameters of the model. The model has learned these specific values from the training data to best approximate the relationship between the features and the price.

To get a more formal definition of these terms, the following article provides a clear, structured explanation.

As the reading mentions, we also have test data. This is a separate portion of our data that the model has never seen during training. We use it to evaluate how well our model performs on new, unseen examples. This is crucial for ensuring our model has learned general patterns, not just memorized the training data. We will cover this in detail in the next lesson.

3. A Practical Walkthrough: The Spam Filter

Let's put all four concepts together in a step-by-step example: building an email spam filter. This should help solidify how these components interact in a real-world scenario.

1. Problem Definition: We want to build a system that can automatically classify an incoming email as either 'Spam' or 'Not Spam'. This is a classification task.

2. Data Collection & Preparation: We gather a large dataset of thousands of emails. For each email, a human has already marked it as spam or not. This becomes our initial dataset.

   • Features: We extract characteristics from each email, such as:
     • Does the subject line contain the word "free"? (Yes/No)
     • Is the sender in our contact list? (Yes/No)
     • The number of exclamation marks in the email. (A number)
     • The number of capital letters used. (A number)
   • Label: For each email, the label is either 'Spam' or 'Not Spam'.

3. Training: We take a large portion of this labeled dataset to be our Training Data. We feed this data to a machine learning algorithm (e.g., a Decision Tree or a Neural Network).

4. Modeling: The algorithm processes the training data and learns the patterns. For instance, it might learn rules like: "If the sender is not in the contact list AND the email contains 'free money', it is highly likely to be spam." The collection of all these learned rules and patterns forms the Model.

The article you just read provides an excellent walkthrough of this exact process. Please return to it to see these steps laid out in detail.

Conclusion

In this lesson, we have defined the four fundamental concepts that form the basis of most machine learning tasks. A clear understanding of these terms is essential before moving on to more complex topics.

Key Takeaways:

• Features are the input variables or predictors that describe your data.
• Labels (or targets) are the output values that you want your model to predict.
• Training Data is the collection of examples—each with features and a corresponding label—that you use to teach the model.
• A Model is the mathematical construct that represents the patterns learned from the training data. It is the artifact that makes predictions on new data.

The relationship is simple: you use training data (containing features and labels) to train an algorithm, which produces a model.

Preview of the Next Lesson:
We briefly mentioned the idea of holding back some data for testing. This is a critical step in building reliable AI systems. In our next lesson, we will focus on the learning outcome: "Split a dataset into training and testing sets and explain the rationale." You will learn why this split is necessary and how to do it correctly to ensure the models you build are truly intelligent and not just good at memorizing.