effects of outliers on data mining