An approach toward improvement of ensemble method’s accuracy for biomedical data classification

Abstract

Amidst rapid technological and healthcare advancements, biomedical data classification using machine learning (ML) is pivotal for revolutionizing medical diagnosis, treatment, and research by organizing vast healthcare-related data. Despite efforts to apply single ML models on clean datasets, satisfactory classification accuracy can still be elusive. In such cases, ML-based ensembles offer a promising solution. This paper explores cascaded ensembles as highly accurate methods. Existing cascade classifiers often partition large datasets into equal unique parts, limiting accuracy due to insufficient amount of useful information processed by weak classifiers of all levels of the cascade ensemble. To address this, we propose an improved cascaded ensemble scheme using a different data sampling approach. Our method forms larger subsamples at each cascade level, enhancing accuracy, and generalization properties during biomedical data analysis. Experimental comparisons demonstrate substantial increases in classification accuracy and generalization properties of the improved cascade ensemble.