Comparison between Conventional and Adjusted Mean Probability of Correct Classification for Two Groups Problem: A Preliminary Study

Abstract

This paper describes a new approach to determine classification performance based on the computation and application of margin of error. This procedure revealed that as the proportion of contamination increases, the misclassification rate and the margin of error also increases. On the other hand, if the mean probability of correct classification is approaching the mean of the optimal probability, the margin of error tends to reduce maximally. The upper and lower classification limits enable us to determine the performance of the technique of interest. If the computed mean probability exceed the upper classification limit this indicates that the rate of misclassification is high. In a general note, we are   confident of the classification result based on this approach. This new technique was applied to investigate the performance of the Fisher linear classification analysis, Fisher’s approach based on the minimum covariance determinant and the probability based classification technique. In general, the performance analysis revealed that as the proportion of contamination increases, the misclassification rate increases thereby producing large margin of error. The implication of large margin of error to classification rule is that the adjusted mean probability based on the margin of error will overshot the upper classification limit which indicates high misclassification rate or possibly highly contaminated data set.

Keywords: Classification, Robust, Mean probability, Margin of error

2010 Mathematics Subject Classification:62H99,62M20