Ensemble feature selection via CoCoSo method extended to interval-valued intuitionistic fuzzy environment

Abstract

Feature selection is a crucial step in the process of preparing and refining data. By identifying and retaining only the most informative and discriminative features, one can achieve several benefits, including faster training times, reduced risk of overfitting, improved model generalization, and enhanced interpretability. Ensemble feature selection has demonstrated its efficacy in improving the stability and generalization performance of models and is particularly valuable in high-dimensional datasets and complex machine learning tasks, contributing to the creation of more accurate and robust predictive models. This article presents an innovative ensemble feature selection technique through the development of a unique Multi-criteria decision making (MCDM) model, incorporating both rank aggregation principles and a filter-based algorithm. The proposed MCDM model combines the Combined Compromise Solution (CoCoSo) method and the Archimedean operator within interval-valued intuitionistic fuzzy environments, effectively addressing the challenges of vagueness and imprecision in datasets. A customizable feature selection model is introduced, allowing users to define the number of features, employing a sigmoidal function with a tuning parameter for fuzzification. The assignment of entropy weights in the Interval-valued intuitionistic fuzzy set (IVIFS) environment provides priorities to each column. The method’s effectiveness is assessed on real-world datasets, comparing it with existing approaches and validated through statistical tests such as the Friedman test and post-hoc Conover test, emphasizing its significance in comparison to current methodologies. Based on the results obtained, we inferred that our structured approach to ensemble feature selection, utilizing a specific case of the Archimedean operator, demonstrated superior performance across the datasets. This more generalized methodology enhances the robustness and effectiveness of feature selection by leveraging the strengths of the Archimedean operator, resulting in improved data analysis and model accuracy.