An evolutionary fuzzy genetic and Naïve Bayesian approach for multivariate data classification

Over past few decades, statistical and soft-computing techniques have become an emerging research area for machine learning problems. Fuzzy logic with better generalization capability and rapport with reality is being used in classification problems immensely. In this paper a fuzzy rule based classification system is modeled as a combinatorial optimization problem. Thus the optimization power of Genetic Algorithm has been applied to select a small number of significant fuzzy rules for a compact classification system applied to multivariate dataset classification. The hybrid algorithm is used here to predict cellular localization sites of proteins in yeast. Two different fitness measures were taken to evaluate the rules generated by the algorithm during evolution process. Than the performance of Fuzzy Genetic algorithm with better fitness measure is compared to the performance of Naïve Bayesian approach to accurately detect and classify patterns. The performance analysis includes testing on train as well as test dataset, created from original dataset. It is found that the fuzzy genetic algorithm with the fitness measure based on data mining support-confidence framework performs better compared to other fitness measure but in terms of classification Naïve Bayesian technique is much more reliable.