Improving kernel ridge regression for medical data classification based on meta-heuristic algorithms

Abstract

Kernel ridge regression (KRR) is a type of machine learning approach that integrates ridge regression with the kernel trick. However, the performance of KRR is sensitive to the values of the hyperparameters that characterize the kernel type. There is a large processing cost, memory expense, and low accuracy performance associated with the existing methods for obtaining these hyperparameter values. The development of meta-heuristic algorithms has helped in solving difficult issues. In this paper, the main improvement is included in the pelican optimization algorithm by applying elite opposite-based learning (EOBL) to improve population diversity in the search space for selecting the best hyperparameters. To confirm and validate the performance of the proposed improvement of KRR, 10 publicly available medical datasets were applied. Depending on several assessment criteria, the results demonstrated that the proposed improvement outperforms all baseline methods in terms of classification performance. The proposed approach has provided more than 92 % of overall accuracy in seven datasets. Of the three datasets, it achieved an overall result of 79 % in producing the highest classification accuracy.