Optimal Stacking Method with GA-Based Feature Selection for Multi-Label Classification

This study proposes an optimal stacking method with genetic algorithm (GA) based feature selection for multi-label classification problems. The most prevalent form of ensemble learning is stacking, which combines the results of multiple base classifiers to improve predictive performance. However, stacking requires careful selection of base classifiers and their hyperparameters, and feature selection. We proposed the use of a feature selection algorithm driven by genetic algorithms (GA) to pinpoint a subset of significant features. This subset is intended to enhance the working capability of the stacking model. The effectiveness of the model is assessed across various benchmark multi-label datasets. The performance results are showing that it outperforms several state-of-the-art multi-label classification methods. Additionally, we experiment to make the analysis of the performance of different hyperparameters and ensemble configurations on the impact of the proposed method. Overall, the proposed optimal stacking method with GA-based feature selection provides a promising solution for multi-label classification problems.