Reinforcing the Decision-making Process in Chemometrics: Feature Selection and Algorithm Optimization

With the development of society, applying data mining schemes in the chemometrics discipline is increasing rapidly which makes this field very popular. However, machine learning algorithms face challenges in selecting best algorithm's parameters as well as selecting the features of the data that affect the decision-making process. Limited studies have in-depth explored ways of enhancing decision support systems in the chemometrics domain. Therefore, this study aims at reinforcing the decision-making process through proposing a robust approach: "feature selection" and "algorithm optimization" in conjunction with "cross-validation". Precisely, stratified tenfold cross-validation method was utilized to evaluate the parameter selection of both Multilayer perceptron and Partial least-squares regression algorithms, from the one hand, and to select the best prediction features, from the other hand. Results exhibited that Multilayer perceptron model overperformed partial least-squares regression model. This confirms that Multilayer perceptron can be efficiently used in the chemometrics discipline. Our result also listed the selected feature for the utilized data. Consequently, current study opens the door for enhancing the industry, generally, and the chemometrics-related manufacturing, especially. It also sheds some light on the significance of adopting cross-validation for model selection and parameter optimization in the chemometrics domain for improving the quality of the decision-making process.