An Evaluation of the Robustness of MTS for Imbalanced Data

In classification problems, the class imbalance problem will cause a bias on the training of classifiers and will result in the lower sensitivity of detecting the minority class examples. The Mahalanobis-Taguchi System (MTS) is a diagnostic and forecasting technique for multivariate data. MTS establishes a classifier by constructing a continuous measurement scale rather than directly learning from the training set. Therefore, it is expected that the construction of an MTS model will not be influenced by data distribution, and this property is helpful to overcome the class imbalance problem. To verify the robustness of MTS for imbalanced data, this study compares MTS with several popular classification techniques. The results indicate that MTS is the most robust technique to deal with the classification problem on imbalanced data. In addition, this study develops a "probabilistic thresholding method" to determine the classification threshold for MTS, and it obtains a good performance. Finally, MTS is employed to analyze the radio frequency (RF) inspection process of mobile phone manufacturing. The data collected from the RF inspection process is typically an imbalanced type. Implementation results show that the inspection attributes are significantly reduced and that the RF inspection process can also maintain high inspection accuracy.