• For training, supervised learning requires a labelled dataset.
• Unsupervised learning identifies hidden data patterns from an unlabeled dataset.
• Reinforcement learning does not require data as it learns by interacting with the environment.
Supervised learning is based on instruction and requires supervision as it follows examples from a labelled dataset to learn.
Unsupervised learning learns by reasoning and does not need supervision as it tries to identify hidden patterns to extract insights from the unlabelled dataset.
Reinforcement learning learns by experience as it explores and exploits different actions to maximize long-term rewards.
Supervised algorithms learn only one type of task based on the labeled dataset. The goal is to predict outcomes for new data belonging to the same domain as the same model need not be applied to a different domain.
Unsupervised algorithm's objective is to gain insights from the unlabeled data that can predict if the new data is part of the cluster or an anomaly.
Reinforcement learning is goal-oriented, and the agent aims to learn sequences of actions by exploration and exploitation in an uncertain environment to maximize future rewards. Reinforcement learning can handle an entirely new scenario it has never encountered.
Supervised learning consists of Classification and Regression. Classification algorithms are applied to detect fraud, spam detection, and classify images, and Regression algorithms help predict sales, house prices, etc.
Unsupervised learning consists of Clustering, Association, Anomaly detection, and dimensionality reduction. Unsupervised learning applications are customer segmentation, market basket analysis, fraud detection, network security analysis, etc.
Reinforcement learning algorithms are either Value-based, Policy-based, or Model-based. Deep Q-Network (DQN), state-Action-Reward-State-Action (SARSA), Asynchronous Advantage Actor-Critic Algorithm (A3C), and Deep Deterministic Policy Gradient (DDPG) are a few Reinforcement algorithms used in Robotics, developing business strategies.
| Criteria |
Supervised ML |
Unsupervised ML |
Reinforcement ML |
|
Definition |
Learns by using labelled data |
Trained using unlabelled data without any guidance. |
Works on interacting with the environment |
|
Type of problems |
Regression and classification |
Association and Clustering |
Exploitation or Exploration |
|
Type of data |
Labelled data |
Unlabelled data |
No – predefined data |
|
Supervision |
Extra supervision |
No supervision |
No supervision |
|
Aim |
Calculate outcomes |
Discover underlying patterns |
Learn a series of action |
|
Algorithms |
Linear Regression, Logistic Regression, SVM, KNN etc. |
K – Means, |
Q – Learning, |
|
Application |
Risk Evaluation, Forecast Sales |
Recommendation System, Anomaly Detection |
Self Driving Cars, Gaming, Healthcare |