ML is broadly split into two primary types — Supervised and Unsupervised learning. Here's a clear breakdown of what each means, when to use them, and how they differ.
Machine learning (ML) is a branch of artificial intelligence that empowers systems to learn from data and make predictions or decisions without explicit programming. ML algorithms analyze data patterns to generate informed predictions or classifications.
📊 ML is broadly categorized into three types: Supervised Learning, Unsupervised Learning, and Reinforcement Learning. This article focuses on the first two.
Common real-world applications include recommendation systems, fraud detection, speech recognition, sentiment analysis, machine translation, self-driving cars, and healthcare diagnostics.
Supervised Learning
Type 1
Supervised Learning
The most popular type of ML. Uses labelled data — historical data that has both the input and the expected output — to train the algorithm. The model learns the input-to-output mapping and eventually takes only the input to predict the output.
Use cases: email spam detection · language translation · predicting house prices
Labelled Data refers to input data paired with the correct output value. For example, in a dataset of 100 houses with their sizes and prices, the "house price" is the labelled data — the answer the model is learning to predict.
Unsupervised Learning
Type 2
Unsupervised Learning
The model is trained on data without labels. Instead of learning from labelled examples, it tries to identify patterns and structures within the input data on its own. The algorithm has to find the pattern — no correct answers are provided.
Use cases: customer segmentation · market analysis · grouping employees
Types of Unsupervised Learning
Clustering
Groups unlabelled input data into "clusters" based on shared characteristics.
Grouping customers by purchasing behavior; patients with similar conditions.
Anomaly Detection
Finds unusual patterns or outliers that don't match expected behavior.
Fraud detection in financial services; suspicious network activity.
Dimensionality Reduction
Reduces the number of input features without losing important information.
Compressing images while preserving features; identifying key stock price factors.
Reinforcement Learning
An agent learns by interacting with an environment and receiving rewards or penalties.
Game-playing AI; robotic control systems. (Covered in a future article.)
Wrapping Up
Supervised learning is your go-to when you have labelled historical data and a clear output to predict. Unsupervised learning shines when you want to discover hidden structure in data with no predefined answers.
Future articles will dive deeper into specific algorithms and techniques within both categories. Happy Learning!