Feature Selection Optimization for Mahalanobis-Taguchi System Using Chaos Quantum-Behavior Particle Swarm

The computational speed in the feature selection of Mahalanobis-Taguchi system (MTS) using standard binary particle swarm optimization (BPSO) is slow and it is easy to fall into the locally optimal solution. This paper proposes an MTS variable optimization method based on chaos quantum-behavior particle swarm. In order to avoid the influence of complex collinearity on the distance measurement results, the Gram-Schmidt orthogonalization method is first used to calculate the Mahalanobis distance (MD) value. Then, the optimal threshold point of the system classification is determined through the receiver operating characteristic (ROC) curve; the misclassification rate and the selected variables are defined; the multi-objective mixed programming model is built. The chaos quantum-behavior particle swarm optimization (CQPSO) algorithm is proposed to solve the optimization combination, and the algorithm performs binary coding on the particle based on probability. Using the optimized combination of variables, a new Mahalanobis-Taguchi metric based prediction system is established to complete the task of precise discrimination. Finally, a fault diagnosis for the steel plate is taken as an example. The experimental results show that the proposed method can effectively enhance the iterative speed and optimization precision of the particles, and the prediction accuracy of the optimized MTS is significantly improved.