Supervised linear estimator modeling (SLEMH) for health monitoring

Abstract

In this research work, the E-Health monitoring system has been developed using fifteen health indicators. These fifteen features were selected by following a Recursive Feature Elimination with Cross-Validation method. The dataset was labeled as per medical limits and segregated into three classes (normal, borderline and onset of unhealthy state). A rigorous process was followed at each step to find out which linear estimator and model is suitable for classifying health condition of persons. Five regression estimators were evaluated and it was found that logistic regression and linear discriminant analysis methods are providing highest accuracy and lowest error for classifying three health states of a patient.