Validation on an Enhanced Dendrite Cell Algorithm using Statistical Analysis

Abstract

Evaluating a novel or enhanced algorithm is compulsory in data mining studies in order to measure it has superior performance than its previous version. In practice, most of studies apply a straightforward approach for evaluation where appropriate performance metrics such as classification accuracy is selected, computes the mean and its variance over several repetitive experiments, and then compares it with the base algorithm or other comparative approach. However, there are limitations using this approach because dataset from different domain tend to produce different error rate thus make their average meaningless as well as susceptible to the outlier. This study demonstrates the mechanism of evaluating an enhanced algorithm using performance metrics and validated it using statistical analysis. In this study, we evaluated the performance of the enhanced algorithm called dendrite cell algorithm using sensitivity, specificity, false positive rate, and accuracy and validated the result using parametric and non parametric statistical significant tests. From the evaluation, the new version of dendrite cell algorithm was statistically proven to have improvement with a significant difference compared to its previous versions in all performance metrics.