Elastic net–guided NSGA-II for scalable multi-objective feature selection in high-dimensional data analytics

Abstract

Redundant, irrelevant, and noisy features make it very hard to analyse high-dimensional data, especially when the number of features is much larger than the number of samples. Conventional feature selection methods, such as filter, wrapper, and embedded methods, are unable to balance predictive accuracy, feature subset compactness, and stability. Multi-objective evolutionary algorithms such as NSGA-II can optimise multiple objectives simultaneously but suffer from high time complexity, unguided search behaviour, and instability in feature subset selection. These limitations highlight a requirement for a more precise and efficient feature selection framework. To tackle those challenges, this study first proposes an Elastic Net–Guided NSGA-II framework that integrates sparsity-aware statistical regularisation and multi-objective evolutionary optimisation. Attributes of Elastic Net coefficients can guide the initialisation of a population and evolution operators, allowing the search space to be explored in an informed manner. And then a stability-aware objective is incorporated to have stable feature subsets across different runs. Utilising several benchmark and high-dimensional biomedical datasets, the experimental results demonstrate that the proposed method consistently outperforms all baseline approaches. It outperforms the classification accuracy of previous state-of-the-art methods by up to 3–5% and reduces the feature subset size by 40–60%, while considerably improving stability, as illustrated in one of the three feature analyses: classification accuracy and feature subset size. In the parity-front-based analysis, it is also shown that solution convergence and diversity have improved. The framework introduced in this paper provides a solution for interpretable, scalable, and easy-to-use high-dimensional feature selection, supporting a wide range of real-world applications, such as bioinformatics, healthcare, and financial analytics. The implementation of the proposed framework is publicly available at https://github.com/subhani6868/ElasticNet-NSGAII-FS.