MOELIGA: a multi-objective evolutionary approach for feature selection with local improvement

2026-03-24 · Source: cs.NE updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, extended

Summary

MOELIGA is a novel multi-objective genetic algorithm designed for feature selection in machine learning. It incorporates an evolutionary local improvement strategy, where subordinate populations refine feature subsets, and employs a crowding-based fitness sharing mechanism for diversity. The algorithm optimizes three objectives: Unweighted Average Recall (UAR) for classification performance, a sigmoid-transformed cardinality ratio for feature subset size, and a geometry-based metric for classifier independence. Experimental results across 14 diverse datasets, including synthetic, real-world, handwritten digit, and cancer-type classification tasks, demonstrate MOELIGA's ability to identify smaller feature subsets while maintaining superior or comparable classification performance against 11 state-of-the-art methods. This approach effectively balances accuracy and dimensionality in complex, high-dimensional scenarios.

Key takeaway

For research scientists developing machine learning models in high-dimensional domains, MOELIGA offers a robust feature selection approach. You should consider integrating multi-objective evolutionary algorithms with local improvement strategies to simultaneously optimize classification performance, model compactness, and classifier independence, potentially yielding more generalizable and interpretable models with fewer features.

Key insights

MOELIGA refines feature selection using multi-objective genetic algorithms with local improvement and diverse fitness sharing.

Principles

• Multi-objective optimization balances competing goals.
• Local search refines globally identified solutions.
• Diversity mechanisms prevent premature convergence.

Method

MOELIGA uses a multi-objective genetic algorithm with subordinate populations for local improvement, optimizing Unweighted Average Recall, sigmoid-transformed feature count, and a geometric classifier independence metric, while employing crowding-based fitness sharing.

In practice

• Use Unweighted Average Recall for imbalanced datasets.
• Apply sigmoid transformation to emphasize smaller feature subsets.
• Employ nearest neighbor distances for classifier-independent feature evaluation.

Topics

• Feature Selection
• Multi-objective Optimization
• Genetic Algorithms
• Evolutionary Local Improvement
• Classifier Independence

Code references

• ldvignolo/moeliga

Best for: Research Scientist, AI Researcher, AI Scientist, Machine Learning Engineer

Related on AIssential

• Multi-population Diversity-guided Genetic Algorithm for Feature Selection in Network Intrusion Detection — MPDGGA improves NIDS feature selection by combining multi-population evolution with diversity-guided operators for better accuracy and feature compactness.
• Towards Automated Knowledge Transfer in Evolutionary Multitasking via Large Language Models — LLMs can autonomously design efficient and effective knowledge transfer models for multi-task optimization.
• MinShap: A Modified Shapley Value Approach for Feature Selection — MinShap adapts Shapley values for robust feature selection by focusing on minimum marginal contributions across permutations.
• IEEE Transactions on Evolutionary Computation, Volume 30, Issue 2, April 2025 — Evolutionary algorithms and multiobjective optimization are advancing diverse fields from logistics to AI design.
• Beyond Cross-Validation: Adaptive Parameter Selection for Kernel-Based Gradient Descents — This paper proposes a novel Hybrid Selection Strategy (HSS) for Kernel-Based Gradient Descent (KGD) algorithms, which integrates bias-variance analysis with the splitting method for adaptive parameter selection.
• Diversity-Driven Offline Multi-Objective Optimization via Nested Pareto Set Learning — Offline multi-objective optimization can achieve diverse, high-quality Pareto sets by mitigating OOD issues and learning nested preferences.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.NE updates on arXiv.org.