
203Forensics Management
ALGORITHMIC CLASSIFICATION
Random forests tend to be very stable in model building. Their relative insensitiv-
ity to the noise that breaks down single decision tree induction models makes them
compare favorably to boosting approaches while they are generally more robust
against the effects of noise in the training dataset. This makes them a favorable
alternative to nonlinear classiers like articial neural nets and support vector
machines.
Each decision tree in the forest is constructed using a random subset of the
training dataset using the techniques of bagging (replacement). A number of
entities will thus be include ...