Evaluation of the Degree of Separation between Two Data Populations with Statistical Algorithms.

摘要

This report investigates various existing statistical methods available in the literature and new statistical methods for evaluating the degree of separation between two peaks or two populations of data. The algorithms evaluated in this study include the direct percentage of overlap between the two populations of data, the Kolmogorov-Smirnov (K-S) test, the area between the receiver operating characteristic (ROC) curve and diagonal line, and the ROC curve length (LROC). These algorithms are compared to the standard reference probability distribution for each data profile to be separated. Evaluations of the algorithms are presented in order to determine the relative degree of separation between the two peaks or distributions. The LROC is determined to provide the best estimation of the degree of separation compared to the other methods.