Feature Selection Method for Gear Health Indicator Using MIC Ranking

In the construction of health indicator for electromechanical equipment, selecting features that exhibit monotonicity, trend characteristics, and a strong correlation with equipment health is paramount to accurately reflect these indices. With the advent of numerous libraries and models for time-series data feature extraction, the range of potential features has expanded significantly. Despite this proliferation, there is a lack of extensive research on effective feature selection. This paper investigates the efficacy of the Maximum Information Coefficient (MIC) method in extracting features that align with the monotonicity and trend-related requirements of electromechanical equipment health indicator. Our experiments indicate that the MIC method adeptly identifies pertinent features for the construction of these indices, underlining its utility in the field of health monitoring for electromechanical systems.