find_outliers

Finds outliers in your data - parameter free.

The algorithm assumes unimodal, non-skewed, but possibly non-normal and correlated dataset of an arbitrary dimension. Outliers are the data points which have less than 5% probability of belonging to the dataset. The approach is empirical, based on simulating 95% quantile of Pearson distributions with zero skew and kurtosis varying from 1.8 (uniform distribution) to 6 (Laplace distribution). The simulation results were linearly fit vs the number of datapoints and the dataset's kurtosis. Dependence of the results on the dataset dimensionality was very slight and was ignored.