Investigation on different learning techniques for weighted kernel regression in solving small sample problem with noise

Abstract

Previously, weighted kernel regression (WKR) for solving small samples problem has been reported. The enhanced WKR has also been proposed with focus on the improvement of the learning techniques and learning functions. In the existing study, the investigation of the learning techniques and learning functions of the WKR is investigated separately. Hence, in this study, one particular learning function that is appropriate to address noisy samples problem with several learning techniques is employed in one investigation framework. The employed learning function is defined as closed form solution function which consists of the error term and regularization term in L2-norm term. For this purpose, an iteration technique, a Ridge Regression (RR) and genetic algorithm (GA) are used as learning techniques to estimate the weight parameters of the WKR for the formulated learning function. Through a number of computational experiments, it is found that RR and GA can offer better prediction accuracy as compared to iteration technique when solving noisy and small sample problems.