Local Outlier Factor

Algorithm overview

Local Outlier Factor (LOF) compares a sample's local reachability density with the densities of its nearest neighbors. It is useful when anomalies live in sparse local regions even though the dataset has several clusters with different densities.

JavaScript implementation

Neighbors.LocalOutlierFactor runs density-based anomaly detection in browser or Node.js code. Training-set detection and novelty detection are deliberately separate, matching sklearn's two usage modes.

Quick start example

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

const X = [[0, 0], [.1, .2], [.2, .1], [1, 1], [8, 8]];
const labels = new Neighbors.LocalOutlierFactor({ nNeighbors: 2, contamination: .2 })
  .fitPredict(X);

const novelty = new Neighbors.LocalOutlierFactor({ nNeighbors: 2, novelty: true });
novelty.fit(X);
console.log(novelty.predict([[.15, .1], [20, 20]]));

Detailed API reference

The constructor accepts nNeighbors?: number, contamination?: number | 'auto', and novelty?: boolean. With novelty: false, call fitPredict(X) and read negativeOutlierFactor. With novelty: true, call fit(X) followed by scoreSamples(X), decisionFunction(X), or predict(X). The fitted threshold is available as offset.