Impact of Feature Selection on the Performance of Classification Algorithms in Predicting Industrial Robot Failures

Industrial robots enhance manufacturing efficiency, productivity, and precision. However, failures can disrupt production lines, leading to losses and significant system impact. In this study, robot failures are predicted using the UR3 CobotOps dataset and the impact of feature selection on the performance of various classification algorithms in predicting two targets (protective stops, and grip losses) is explored. Initially, the baseline performance of classifiers without feature selection has been evaluated. Then, two different feature selection methods (recursive feature elimination and chi-square) are applied to select the top 10 features and reassess the classifier’s performance. High classification success rates are obtained with Decision Tree and Random Forest after feature selection in this study, which tests five different classifiers (Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, and k-Nearest Neighbors) in the classification stage. This paper provides valuable insights into the different applications of classifiers, contributing to the field of machine learning by identifying different feature selection techniques and their impacts on classification accuracy. According to the experimental tests, an accuracy rate of about 99% has been obtained when Random Forest is used. This success has been also achieved when Chi-Square is used for feature selection. This paper shows that this prediction can be achieved in a shorter time using feature selection.