Improving Classifier Training Efficiency with Genetic Algorithm Feature Selection

A genetic algorithm feature selection technique is used to improve efficiency of a classifier training process while maintaining its output quality. To achieve this goal, values of some hyper-parameters in the genetic algorithm for feature selection are varied and investigated. Also, different focuses of fitness functions are explored. A set of experiments is conducted on two types of data, unknown and known feature importance. Training time and average accuracy score of the result classifier from each experiment are collected and compared. The results show that there can be an improvement in classifying time and quality of the classifier when appropriate hyper-parameters and fitness function are combined.