An ensemble-based framework for biomedical classification problems

Model selection is an essential step when applying machine learning to classification problems. It is typically carried out by the practitioner who strives to identify the most suitable classifier for a given problem. Given the variety of classifiers available and the difficulty in predicting which one will yield the best performance depending on the characteristics of the problem, this is by no means a simple task. Biomedical problems pose a significant challenge in this regard due to their numerous data intrinsic characteristics that are known to impair classification performance. Given that different classifiers perform well for different biomedical problems, combining them into an ensemble would seem practical. However, the practitioner still needs to determine how to combine them. This paper presents an ensemble-based framework that automates the training and combination of different classifiers in order to relieve practitioners of this burden whilst obtaining highly competitive performance. The effectiveness of the proposed framework was evaluated on several biomedical problems from the literature.