Analysis of Nine Instance-Based Genetic Algorithm Classifiers Using Small Datasets

Abstract

The application of genetic algorithm (GA) has emerged covering various areas including data classification. In data classification, most studies of GA were focused on the enhancement of GA and development of different types of GA classifiers. To the best of our knowledge, there is no study has been conducted to examine the influence of GA operators based on the size of data set towards training time and generalization ability. Therefore, this study develops and compares nine Instance-based genetic algorithm (IbGA) classifiers with different combinations of GA operators. The goal of this comparison is to examine and identify the best combination of GA operators which have performed better on generalization ability and training time efficiency. Nineteen benchmark data sets were used in this study. The non-parametric statistical tests were applied to justify the comparison results. The statistical tests suggest that the combination of roulette wheel selection and uniform crossover operator is the best combination of IbGA model although the training time is a bit lengthier than compared to other IbGA models.