Anomaly Detection: Statistical and Learning-Based Methods

Module: Machine Learning | Difficulty: Advanced

Isolation Forest

Anomalies are easier to isolate. Path length :

where .

One-Class SVM

Autoencoder

Comparison

import numpy as np
from sklearn.ensemble import IsolationForest

class AutoencoderAnomaly:
    def __init__(self, encoder, decoder, threshold_percentile=95):
        self.encoder = encoder; self.decoder = decoder
        self.threshold_percentile = threshold_percentile
    def fit(self, X):
        recon = self.decoder(self.encoder(X))
        errors = np.mean((X - recon)**2, axis=1)
        self.threshold = np.percentile(errors, self.threshold_percentile)
    def predict(self, X):
        recon = self.decoder(self.encoder(X))
        errors = np.mean((X - recon)**2, axis=1)
        return errors > self.threshold

Research Insight: Isolation forest works because anomalies have shorter average path lengths in random trees. The key insight is that anomalies are few and different, so they are isolated faster than normal points.