Combining Correlation Technique with Exhaustive Search Feature Selection Method for Rotating Machinery Fault Diagnosis

The feature selection in classification tasks as a need for diagnosing a fault in rotating machinery exceedingly plays a very notable role in the machine learning framework. Consequently, this study proposes a combination of a correlation technique with exhaustive search as a feature selection method for diagnosing rotating machinery faults. In any case, this method can be called as a hybrid method because it combines between the correlation technique as a filter method and an exhaustive search as a wrapper method. The correlation of each feature with the target of labeling normal and abnormal information is measured by Pearson’s correlation matric, in which abnormal conditions indicate a failure in the rotating machinery. The top five best correlations at this stage are taken as the selected feature in the filtering stage. Based on these features selected, each feature is combined and its performance is considered through the training of a classification model. The combination of features with the highest accuracy is the final selected feature subset. Finally, the proposed method is able to successfully demonstrate the diagnosis of rotating machinery faults with normal and abnormal classification.