Outlier filtering in a sample

Abstract

The problem of filtering outliers in the sample is considered. A genetic algorithm for outlier filtering is proposed, its efficiency is tested on synthetic and real data in the linear regression problem. Synthetic data was generated by applying normally distributed random noise to a linear function. Real data check was performed on The Boston Housing Dataset. Since normally distributed random noise with small variance distorts the original function rather weakly and may, in general, have no outliers, the proposed outlier filtering algorithm showed a noticeably greater efficiency on real data, however, the positive effect of the proposed outlier filtering method was also noticeable on synthetic data.