Comparative analysis of decision tree and support vector machine algorithms for classifying pump impellers

The identification and controlled replacement of defective parts in a machine increases equipment reliability. A pump is a device used to move fluids, and the pump impeller is one of its critical components, as it converts the mechanical energy of the motor into the kinetic energy of the fluid. If the impeller is damaged, the pump may fail. This paper presents a comparative analysis of two machine learning algorithms—decision trees and support vector machines—applied to the classification of pump impellers as "good" or "bad" based on their physical condition, a key task in preventive maintenance. The experiment uses a balanced dataset of impeller images to compare the performance of both algorithms by evaluating metrics such as accuracy, total cost, training time, prediction speed, and the confusion matrix. The main results show a test accuracy of 87.9% using decision trees and 98.7% using support vector machines; however, the latter requires more training time. It is concluded that both algorithms are effective for classifying the component under study. They are computationally efficient but pose challenges such as overfitting and hyperparameter tuning. Additionally, the use of hyperplanes and an optimized kernel in the SVM algorithm makes classification more efficient, although it requires more time and data for training.