Ant Colony Optimization for Markov Blanket-Based Feature Selection. Application for Precision Medicine

摘要

In this work, we address feature subset selection in the case when the variables exert a null or weak marginal effect on the target variable, a situation called "pure" epistasis hereafter. We explore the Markov blanket approach, to tackle epistasis detection, and we introduce SMMB-ACO. This method combines Markov blanket learning with stochastic and ensemble features and guides the stochastic sampling process by incorporating ant colony optimization. We first analyze the impact of parameter adjustment on SMMB-ACO complexity. Then using simulated and real data, we compare SMMB-ACO with four other methods, including its former version SMMB. We show that SMMB-ACO compares well with three state-of-the-art methods and that SMMB-ACO is more stable than SMMB. On the real dataset, the detection ability of SMMB-ACO is close to that of the best approach, which is a slow method, and SMMB-ACO is the fastest algorithm behind a much less performing method.