Apply isolation forest to new samples and flag anomalies

Scores new samples using a fitted isolation forest (fit_isolation_forest_logratio), computes empirical p-values relative to the training-score distribution, and flags samples exceeding the training \((1 - \texttt{fpr\_target})\) quantile.

Usage

apply_isolation_forest_logratio(fit, x_test, fpr_target = 0.05)

Arguments

fit

An isolation_forest_logratio_fit object.

x_test

Numeric matrix (samples \(\times\) features). All training features must be present.

fpr_target

Numeric in (0, 1). Empirical FPR target for the flagging threshold. Default 0.05.

Value

List with score (anomaly score in \([0,1]\); higher = more anomalous), pvalue_emp, threshold, flagged.