Improving Efficiency of the Genetic Algorithm Feature Selection with Dynamic Programming

A dynamic programming algorithm is applied to the genetic algorithm feature selection program to reduce the training time. After each generation, an accuracy score of each genome in a selected population is obtained and stored for later use. An array data structure and conversion of a genome information to an array index is used to achieve the O(1) time complexity when storing and retrieving accuracy score of each genome. The demonstrate the benefit of dynamic programming, training time data from the genetic algorithm feature selection program solving a classification problem with three machine learning models are recorded and compared. The results in one of the machine learning models show significant improvement, especially when the feature size of a data set and population size is large.