Analysis of Feature Selection Methods On Clove Quality Classification

Proper feature selection is becoming increasingly important in the context of clove quality classification in Indonesia; effective feature selection can identify the most relevant attributes of clove quality. This study investigates 3 famous feature selection methods, namely mutual information, recursive feature selection, and lasso regression, for classifying clove quality using various models, i.e. Naïve Bayes, K-NN, and Decision Tree. The study results showed that the mutual information feature selection method impacted the K-NN method by 1.05%; for naïve Bayes and decision trees, the accuracy did not increase. For the recursive feature elimination (RFE) feature selection method, the accuracy of K-NN increased by 2.16%, the decision tree by 0.37%, and naïve Bayes accuracy did not increase. In addition, using Lasso Regression feature selection, the accuracy of naïve Bayes increased by 2.97%, K-NN by 2.28%, and decision tree accuracy decreased.