Enhanced prediction of misalignment conditions from spectral data using feature selection and filtering

This paper proposes a novel method for the use of genetic algorithm-based feature selection and signal filtering to construct reliable calibration models of shaft misalignment. Determination and selection of the key feature(s) is crucial to the predictive performance of calibration models. Even with proper feature selection, the predictive performance of calibration models can be enhanced by filtering the raw spectral data. This improvement results because a filter removes the unwanted variation of predictor variables that is orthogonal to response variables. This is the first work that attempts to develop a systematic calibration model based on genetic algorithm-based feature selection and orthogonal filtering. A case study shows that the proposed calibration model for shaft misalignment conditions produces better predictive performance than traditional multivariate statistical approaches such as principal component regression and partial least squares.