Change point detection aims to identify structural breaks in the probability distribution of a time series. Existing methods either assume a parametric model for within-segment distributions or are based on ranks or distances and thus fail in scenarios with a reasonably large dimensionality.

changeforest implements a classifier-based algorithm that consistently estimates change points without any parametric assumptions, even in high-dimensional scenarios. It uses the out-of-bag probability predictions of a random forest to construct a classifier log-likelihood ratio that gets optimized using a computationally feasible two-step method.

See [1] for details.

[1] M. Londschien, P. Bühlmann, and S. Kovács (2023). "Random Forests for Change Point Detection" Journal of Machine Learning Research