Isolation forest parameters

Number of trees: 100 Subsample size (ψ): 128 Contamination rate: 6%

View

Population

Total claims—

Injected fraud—

Trees built—

Tree height limit—

Live detection score

Mean score — genuine—

Mean score — fraud—

Top-N precision—

Selected claim—

Low anomaly score High anomaly score

Every claim is a point in 3D feature space (claim amount, days since policy start, claim frequency). Colour encodes each claim's live Isolation Forest anomaly score s(x,n) = 2^{-E(h(x))/c(n)}. Drag to orbit, scroll to zoom, click a point to inspect it. The histogram below splits scores by the claim's true (normally hidden) fraud label so you can verify the ensemble is genuinely separating them.