Hands-On Machine Learning with C++

Detecting anomalies with isolation forest

The idea of an isolation forest is based on the Monte Carlo principle: a random partitioning of the feature space is carried out so that, on average, isolated points are cut off from normal ones. The final result is averaged over several runs of the stochastic algorithm, and the result will form an isolation forest of corresponding trees. The isolation tree algorithm then builds a random binary decision tree. The root of the tree is the whole feature space. In the next node, a random feature and a random partitioning threshold are selected, and they are sampled from a uniform distribution on the range of the minimum and maximum values of the selected feature. The stopping criterion is the identical ...