Enhanced data point importance for subset selection in partial least squares regression: A comparative study with Kennard-Stone method

In multivariate data analysis, the selection of representative subsets of samples is crucial for developing accurate predictive models. This study evaluates the application of the Enhanced Data Point Importance (EDPI) method for subset selection, comparing its performance with the widely-used Kennard-Stone algorithm. The EDPI method ranks all the data points using the DPI and layered convex hull approach, resulting in a ranked sequence of points based on their importance in the dataset, with the most important point being the most informative. Both methods were applied to two distinct datasets, and Partial Least Squares Regression (PLSR) models were developed for each subset to assess predictive performance. The EDPI method demonstrated comparable performance to the Kennard-Stone method across various sample sizes. The EDPI-PLS models achieved lower Root Mean Square Error of Prediction (RMSEP) values with fewer samples, indicating efficient subset selection, and the method is less inclined to select the influential points in the dataset. Moreover, the running time analysis highlighted the computational efficiency of the EDPI method, especially in high-dimensional datasets. These findings suggest that EDPI is a robust and informative strategy for sample subset selection, offering advantages in predictive accuracy and computational efficiency.