Improving firefly algorithm-based logistic regression for feature selection

One of the major problems with data classification is when it has a high dimensional, small sample size and contains irrelevant and redundant features. Binary firefly algorithm is one of the nature-inspired metaheuristic algorithms which was designed to solve the discrete optimization problem, such as feature selection. However, the binary firefly algorithm version needs a transfer function that changes search space from continuous to the discrete. In this paper, several transfer functions are investigated and explored. To validate the efficiency, and to which extent that functions impact the results, an extensive experiment was carried out on three chemometrics datasets while the logistic regression method and firefly algorithm were used to classify data and select features. The experimental results show that V2 function has consistency in feature selection and it performs high classification with better performance to the iterations.