Predicting and Managing Medication Adherence Using Random Forest with Light Gradient Boosting Method and Mobile Health Data

In recent years, the importance of medication adherence is increased healthcare industry as prevalence of chronic diseases, like diabetes, hypertension, and asthma has improved. The traditional models failed in providing the correct dosage at correct time and at correct day. So, to overcome these limitations in this research, Random Forest- Light Gradient Boosting Method (RF-LightGBM) is proposed for predicting and managing medication adherence which is employed on Health and Retirement Study and consists of collection of health data which is gathered from the older adults in the United States. These data are pre-processed by using Synthetic Minority Over-sampling Technique - Edited Nearest Neighbours (SMOTE-ENN) and Natural Language Toolkit (NLTK) and optimal features are selected by Recursive Feature Elimination (RFE) that worked as wrapper for eliminating irrelevant attributes. The selected optimal features are classified by RF and LightGBM for distinguishing adherent and non-adherent patients and predicting medication adherence. Finally, RF- LightGBM is combined for associating the strengths of both RF and LightGBM which improved the accuracy and efficiency for classification and prediction. the proposed RF - LightGBM achieved better accuracy of 0.995, specificity of 0.984 and sensitivity of 0.992 when compared with existing Support Vector Machine (SVM).