Abstract:<p>Feature selection (FS) aims to identify the most relevant and non-redundant feature subset for improving the classification accuracy, which is regarded as a NP-hard problem. Some heuristic methods, such as particle swarm optimization (PSO) have achieved great success, however, with the increase of feature quantity, the solution space is too large, resulting in lower search efficiency. Recent discretization-based FS methods map the search of feature domain into cut-point domain, which shrinks the solution space and improve the performances significantly. In this paper, considering the conflicts between different objectives, we proposed an evolutionary multi-objective optimization framework for discretization-based FS. To obtain the Pareto solutions, a flexible cut-point PSO (FCPSO) which can select an arbitrary number of cut-points for discretization is introduced to help better explore the relevant features. In FCPSO, a particle update and a novel adaptive mutation operator are alternatively used to effectively find the relevant features and remove the redundant features. At last, to select the best feature subset, a Pareto ensemble method is designed to generate a number of feasible solutions based on Pareto set followed by a hierarchical solution selection process. We implemented the proposed framework by using three representative multi-objective evolutionary algorithms and compared them with some state-of-the-art methods. Experimental results on ten benchmark microarray gene datasets demonstrate that our proposed framework significantly outperforms other methods in terms of test classification accuracy with a competitive size of feature subset.</p>

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to perform feature selection (FS) in high - dimensional data in order to improve the accuracy of classification tasks. Specifically, the paper focuses on how to maintain or improve classification performance while reducing the number of features. Since feature selection is an NP - hard problem, as the number of features increases, the solution space becomes very large, resulting in reduced search efficiency. To solve this problem, the paper proposes a multi - objective evolutionary optimization framework based on discretization. ### Main contributions of the paper: 1. **Modeling of multi - objective optimization problems**: The paper models the feature selection problem as a multi - objective optimization problem (MOP), considering three objectives: classification error rate, proportion of selected features, and distance measure between samples. 2. **Flexible cut - point selection strategy**: A flexible cut - point particle swarm optimization algorithm (FCPSO) is proposed. This algorithm can select any number of cut - points for each feature, thus better exploring relevant features. 3. **Solution selection in Pareto set**: A Pareto set method is designed, combined with a hierarchical solution selection process, to select the best feature subset from multiple feasible solutions. 4. **Systematic comparative experiments**: The proposed framework is implemented under different multi - objective evolutionary algorithm frameworks (such as MOEA/D, NSGA - III, and SPEA - II) and compared with existing advanced methods to verify its effectiveness and superiority. ### Specific problem description: - **Challenges of feature selection**: In high - dimensional data, there are many irrelevant or redundant features. These features will affect the performance of classification algorithms, leading to over - fitting problems and reducing classification accuracy. - **Limitations of existing methods**: Traditional feature selection methods (such as sequential selection or feature - ranking - based methods) are prone to fall into local optima and cannot fully consider the interaction between features. - **Importance of discretization**: Discretization can convert continuous values into discrete values, which helps to remove noise and improve the performance of classifiers. ### Solutions: - **Multi - objective optimization framework**: Find a balance point by simultaneously optimizing the classification error rate, proportion of selected features, and distance measure between samples. - **Flexible cut - point selection**: The FCPSO algorithm allows any number of cut - points to be selected for each feature, improving the flexibility and efficiency of the search. - **Pareto set method**: Generate multiple feasible solutions through the Pareto set method and combine with the hierarchical solution selection process to finally select the best feature subset. ### Experimental results: - **Performance improvement**: The experimental results show that the proposed framework significantly outperforms other methods on multiple benchmark datasets, especially in terms of testing classification accuracy. - **Feature subset size**: While maintaining competitive classification accuracy, the proposed framework can select fewer features. In conclusion, this paper effectively solves the feature selection problem in high - dimensional data and improves the performance of classification tasks by introducing multi - objective optimization and flexible cut - point selection strategies.

An evolutionary multi-objective optimization framework of discretization-based feature selection for classification

Many-objective optimization of feature selection based on two-level particle cooperation

Feature subset selection via an improved discretization-based particle swarm optimization

A particle swarm optimisation based multi-objective filter approach to feature selection for classification

An evolutionary multiobjective method based on dominance and decomposition for feature selection in classification

A multi-objective feature selection approach based on chemical reaction optimization

A Multi-objective Feature Selection Method Considering the Interaction Between Features

Multi-objective particle swarm optimization with adaptive strategies for feature selection

A Multi-Objective Particle Swarm Optimisation for Filter-Based Feature Selection in Classification Problems

MPF-FS: A multi-population framework based on multi-objective optimization algorithms for feature selection

A problem-specific non-dominated sorting genetic algorithm for supervised feature selection

Differential Evolution Based Feature Selection: A Niching-based Multi-objective Approach

Multi-objective optimization algorithm based on clustering guided binary equilibrium optimizer and NSGA-III to solve high-dimensional feature selection problem

Multi-Objective Self-Adaptive Particle Swarm Optimization for Large-Scale Feature Selection in Classification

Multiobjective Differential Evolution for Feature Selection in Classification

A Constrained Competitive Swarm Optimizer With an SVM-Based Surrogate Model for Feature Selection

A Surrogate-Assisted Evolutionary Feature Selection Algorithm with Parallel Random Grouping for High-Dimensional Classification

Feature selection using feature ranking, correlation analysis and chaotic binary particle swarm optimization

Surrogate Sample-Assisted Particle Swarm Optimization for Feature Selection on High-Dimensional Data

A self-adaptive multi-objective feature selection approach for classification problems

Feature selection using bare-bones particle swarm optimization with mutual information