Multiple indefinite kernel learning for feature selection

This study proposes a novel approach to feature selection through the integration of multiple indefinite kernel learning algorithms. Leveraging the strengths of Support Vector Machine (SVM), ElasticNet-SVM, Random Forest Multiple Kernel Learning (RFMKL), Generalized Multiple Kernel Learning (GMKL), Incremental Kernel Principal Component Analysis (IKPCA), Structured Support Vector Machine (SSVM), SSVM-P, and Mutual Information Kernel Feature Selection (MIK-FS), our method aims to identify the most informative features while mitigating the curse of dimensionality. By employing a diverse set of algorithms, our framework explores the data from various perspectives, enhancing the robustness and adaptability of feature selection. SVM and ElasticNet-SVM ensure robust classification performance, while RFMKL and GMKL enable effective integration of heterogeneous data sources. IKPCA aids in dimensionality reduction, while SSVM and SSVM-P offer structured output prediction capabilities. MIK-FS provides a principled approach for selecting features based on mutual information. Through comprehensive experiments on diverse datasets, we demonstrate the efficacy of our approach in achieving superior feature selection performance, outperforming state-of-the-art methods in terms of classification accuracy and computational efficiency.

Keywords: SVM, ElasticNet-SVM, RFMKL, GMKL, IKPCA, SSVM, SSVM-P, MIK-FS, feature selection