## TL;DR
Anomaly detection identifies data points that deviate significantly from expected patterns — critical for fraud detection, industrial monitoring, and cybersecurity.

## Core Explanation
Three paradigms: supervised (labeled anomalies, rare class problem), semi-supervised (labeled normal data only, one-class classification), unsupervised (no labels, assumes anomalies are rare and different). Key methods include statistical (z-score, IQR), distance-based (kNN, LOF), and deep learning approaches (autoencoders, GANs).

## Detailed Analysis
One-class SVM learns a decision boundary enclosing normal data with maximum margin to origin. Deep SVDD finds the smallest hypersphere enclosing normal representations. In time series, LSTM-based predictors flag points with unexpectedly high prediction error as anomalies.

## Further Reading
- PyOD: Python Outlier Detection Toolkit
- scikit-learn: Novelty and Outlier Detection
- Papers With Code: Anomaly Detection