Improving the Performance of Robust Partial Least Squares Regression Using an Iterative Approach

Abstract

The robust partial least squares regression method provides a solution to noise and outliers in the estimated models by maximizing the explanation ratios of the independent and dependent variables (determination coefficient). Three proposed methods were presented. The first depends on the iterative method, which determines outliers and estimates them using the initial estimated values and the mean square error, as well as determining the optimal value that gives the least mean squares error for the partial least squares regression model. The second (Robust-Iteration) and third (Iteration-Robust) proposed methods rely on hybrid estimators of the iteration and robust approaches, that maximize the explanation ratios in the independent and dependent variables while minimizing the mean squared error. Simulation results and real data from chemical experiments (the quality of a chemical product based on various physicochemical properties) demonstrated the efficiency and accuracy of the proposed methods in handling outliers and noise in the data compared with the partial least squares regression method.