Comparative assessment of soft computing and SVM architectures for multi-class automobile engine fault classification

The advancing complexity of automobile structure and continuous evolvement of automobile functionality has increased the difficulties of automobile fault diagnosis. This study investigates the effectiveness and efficiency of particle swarm optimization-based adaptive neurofuzzy inference system (PSO-ANFIS) models in offline automobile fault diagnosis. The results of this model were compared with error-correcting output code (ECOC) support vector machines with one-vs-one (ECOC-SVM-OVO) and one-vs-all (ECOC-SVM-OVA) based structures, probabilistic neural network (PNN), and standalone adaptive neurofuzzy inference system (ANFIS). Fifty cars were diagnosed, symptoms were classified into ten (10), and coded in binary forms, while likely faults were grouped into 10. The models were evaluated against relevant classification performance metrics and computational time. Similar performance metrics were obtained for the ECOC-SVM-OVO, ECOC-SVM-OVA, PNN and PSO-ANFIS models (accuracy = 1, error = 0, specificity = 1, false positive rate = 0, kappa statistic = 1). The standalone ANFIS model performed the least (accuracy = 0.75, error = 0.25, specificity = 0.97, false positive rate = 0.03, kappa statistic = 0.21), though at the least computational time (2.57 s). Although standalone ANFIS and PSO-ANFIS models could be used as classification models, their efficiencies and effectiveness are lower than those of the PNN and SVM architectures in this study.