Multi-objective Optimization for Feature Selection: A Review

A multi-objective optimization problem (MOP) refers to a methodology employed to identify optimal solutions for issues that fall under the category of NP-hard. This method yields a collection of compromise answers rather than a single optimal answer. Feature selection serves as a critical preprocessing phase in machine learning aimed at enhancing the effectiveness of learning strategies by eliminating features unrelated or redundant to the input. The MOP approach is being employed on multiple occasions to solve the feature selection issues, yielding notable outcomes. This chapter reviews the presented feature selection methods focused on the MOP approach. To this end, we first introduce the MOP and then discuss the issue of feature selection and its necessity in machine learning. Next, we will examine the feature selection approaches introduced thus far, which are based on the MOP approach. Then, we will explore these feature selection methods from the supervision perspective (Supervised, Unsupervised, Semi-Supervised) and interaction with the learning algorithm (Filter, Wrapper, Hybrid, and Embedded). Ultimately, we will discuss the challenges and issues that can still be solved using the ACO algorithm in feature selection.