Feature selection for healthcare study: Modified WSM and Machine Learning approach

Over a past decade it has been observed that a good research is done in healthcare using Machine Learning(ML) and Deep Learning(DL). To carry on the research or implementation using ML or DL, the main concern is database fields or feature selection for the implementation to be done based on the disease or diagnosis method. This paper mainly focus on the two methodologies for feature selection, the first method is a survey based method with Modified Weighted Sum Method (MWSM) implemented on the data collected and second is the feature selection method using machine learning. The said methods are applied to the data collected from diabetic patients and doctors for finding the features useful for diagnosing Hypoglycemia state (lowering of blood glucose level(BGL) in diabetic patients. Comparative analysis is shown for these method and thus concluding for the features selected. Features considered for this work were superficial body features for hypoglycemia detection. The features which we selected from modified WSM were used for feature selection algorithm implementation using machine learning. Univariate Statistical Tests(UST), Recursive Feature Elimination(RFE), Principle Component Analysis(PCA) and Feature Importance(FI) algorithm were implemented on data thus obtained resulting in omitting or rethinking for features.