Types of Machine Learning: Supervised, Unsupervised and Reinforcement

Hello students, welcome to Kwickprep. A child learns from a teacher, from exploring alone, and from praise or scolding. Machines learn in the same three ways. Today we learn supervised, unsupervised and reinforcement learning, with a clear example of each.

First, a quick meaning. Machine learning, or ML, is a part of AI where computers learn from data. The programmer does not write every rule step by step. Instead, the model finds patterns in data and gets better with more experience.

The first type is supervised learning. Here the model learns from labelled data. A label is the correct answer attached to each example, like a photo marked cat or dog. It is like a student practising with an answer key, where a teacher supervises. Supervised learning has two kinds, classification and regression.

Here is labelled data for an email spam filter. The first email says you won ten lakh rupees, and it is labelled spam. A message about a class test on Monday is labelled not spam. Claim a free recharge is labelled spam. A fee receipt is labelled not spam. After seeing thousands of such examples, the model can label a new email by itself.

Supervised learning answers two kinds of questions. Classification predicts a category, like spam or not spam, or pass or fail. Regression predicts a number, like the price of a house or tomorrow's temperature.

The second type is unsupervised learning. Here the data is unlabelled, which means no correct answers are given. The model finds hidden structure or groups on its own. The most common method is clustering, which puts similar items into the same group. Another is association, which finds items that often appear together, like bread and butter in a shopping cart.

Here is data from a kirana store app, with no labels at all. Customer A visits twelve times a month and spends three hundred rupees. Customer B is very similar, with eleven visits and two hundred fifty rupees. Customer C visits only twice but spends four thousand rupees. Customer D is similar to C. So the model makes two clusters: frequent small buyers, and rare big buyers. Nobody told it these group names.

The third type is reinforcement learning. Here a learner called an agent tries actions, one after another. The world around it, called the environment, gives a reward for a good action and a penalty for a bad one. The agent's goal is to collect the most reward over time. So it learns by trial and error, without any answer key.

Let us follow the reward loop step by step. The agent first sees the state, which is the current situation. Then the agent takes an action. Because of that action, the environment changes. The environment sends back a reward or a penalty. The agent updates what it has learned and repeats the loop many thousands of times.

Imagine a program learning to solve a maze game. The agent is the program that plays. Its actions are moving up, down, left or right. It gets plus ten points for reaching the exit. It gets minus one point each time it hits a wall. After many games, it learns the shortest safe path. The same idea helps computers learn to play chess and to control robots.

Let us compare the three types side by side. Supervised learning uses labelled data and learns from correct answers. Unsupervised learning uses unlabelled data and learns from hidden patterns. Reinforcement learning starts with no dataset at all, and learns from rewards it earns while acting.

Board exams often ask for one example of each type, so learn these. Supervised learning: an email spam filter trained on labelled emails. Unsupervised learning: grouping customers of a shop by buying habits. Reinforcement learning: a program that learns a game by earning points.

Here are three cases to think about. First, sorting news articles into topic groups, with no labels given. Second, predicting pass or fail from past results that already show who passed. Third, a car in a simulator rewarded for staying in its lane. Pause the video and name the type for each case.

The answers are unsupervised, supervised classification and reinforcement learning, so let us revise. Supervised learning uses labelled data, like practising with an answer key. Classification predicts a category, and regression predicts a number. Unsupervised learning uses unlabelled data and finds clusters by itself. Reinforcement learning has an agent that learns from rewards and penalties. Remember one example of each: a spam filter, customer groups, and a game playing program.