Kernel ridge regression improving based on golden eagle optimization algorithm for multi-class classification

Abstract

        Kernel Ridge Regression unites machine-learning supervision with ridge regression principles through application of the kernel trick. The method proves most effective when dealing with regression problems that exhibit non-linear input-output relationships. The kernel trick enables Kernel Ridge Regression (KRR) to execute ridge regression. The technique acquires knowledge of non-linear functions in a high-dimensional space through the regularization methods of ridge regression. The effectiveness of KRR depends on the hyper-parameter settings which determine the kernel type. The current methods for obtaining hyper parameter values face three major challenges: high processing costs, substantial memory requirements and poor accuracy levels. The research presents an important advancement to the golden eagle optimization framework. The enhancement adds elite opposite-based learning (EOBL) to boost population diversity throughout the search space. We implement this method to effectively select the best hyper parameters. The combination of EOBL with KRR can yield to improve predictive accuracy. By choosing the elite solutions and including the opposition-based methodologies, the model can avoid local optima and expand the range of the potential solution, which should yield better results in the contexts of complex data sets. The proposed Kernel Ridge Regression enhancement was tested using ten multi-class datasets available to the public to prove its efficacy. The results from multiple assessment criteria demonstrated that the proposed enhancement achieved superior categorization efficacy compared to all basic procedure techniques.