Simple nonparametric tests for change point detection in time series

The paper describes new simple non-parametric tests for recognizing change points in a time series separating fragments obeying different distributions. The test is based on Hill’s assumption A(n) and the generalized scheme of the Bernoulli scheme. To identify the change point, it is proposed to use nonparametric tests that use the effect of stabilizing the relative frequency of success in the Bernoulli scheme. Both samples without repeated elements and samples with repeated element are allowed. The test allows both online and offline implementations without requiring significant computing power. Experiments demonstrate the stability and sensitivity of the proposed tests. The tests recognize change points between two distributions with high accuracy, both in the case of different location parameters and in the case of different scale parameters.