we explain why usingsmall sub-samples brings about better isolation models and examine the changesof detection behaviour by adjusting the evaluation height-limit.
anomaly detection using iForest is a twostage process. The first (training)stage builds isolation trees using subsamples of the given training set. The second (evaluation) stage passes test instances through isolation trees to obtain an anomalyscore for each instance.
4.1 Training Stage
In the training stage, iTrees are constructed by recursively partitioning a sub-sample X′until all instances are isolated. Details of the training stage can befound in Algorithms 1 and 2. Each iTree is constructed using a sub-sample X′randomly selected without replacement from X, X′ ⊂X.
4.2 Evaluation Stage
