Performance Analysis of ML Algorithms for Brake System Failure Detection

---

Abstract

Hydraulic brakes play a pivotal role in ensuring passengers, drivers and goods safety in vehicles making them indispensable components. Therefore, monitoring and predicting failure of brake becomes very important. Machine learning (ML) based algorithms are used for predicting the failure of the braking system. In this paper, we have carried out performance analysis of almost all major algorithms (ML classifiers) to predict and monitor the brake failure. We used dataset from Scania Trucks in these algorithms and their performances are evaluated using the parameters, sensitivity, specificity and accuracy. Almost all algorithms give accuracy around 98%, except Naïve Bayes having 92% accuracy. It has been found that the stochastic gradient descent (SGD) algorithm yields better performance across all parameters: sensitivity 97.709%, specificity 96.146%, and accuracy 96.925%.

---

Keywords

Air pressure system (APS), Feature selection, Brake failure monitoring, PCA, Gradient Boosting Classifier, Support Vector Machine (SVM), Decision Tree, Random Forest, KNN Classifier.