EllipticEnvelope anomaly detection

Algorithm overview

EllipticEnvelope fits a robust Gaussian-shaped envelope around inliers and flags samples whose robust Mahalanobis distance exceeds a contamination-based threshold. It works best when the inlier cloud is approximately elliptical.

JavaScript implementation

Covariance.EllipticEnvelope wraps MinCovDet, so JavaScript applications can score or label tabular anomalies locally in a browser or Node.js process.

Quick start example

import { Covariance } from '@kanaries/ml';

const X = [[0, 0], [.1, .2], [.2, .1], [-.1, 0], [8, 8]];
const detector = new Covariance.EllipticEnvelope({ contamination: .2, randomState: 42 });
detector.fit(X);
console.log(detector.predict(X)); // 1 for inlier, -1 for outlier

Detailed API reference

The constructor accepts contamination in (0, 0.5] plus all MinCovDet options. Methods are fit(X), scoreSamples(X), decisionFunction(X), and predict(X). The fitted offset, location, and covariance are readable properties.