Boosting with Feature Selection Technique for Enhancing The Prediction of Depression and Suicide Risk in Elderly

Abstract

The objective of this research is to analyze the questions that important to predict the depression and suicide risk in elderly by using boosting algorithms (AdaboostM2) with minimum Redundancy Maximum Relevance (mRMR) feature selection. Data were collected from 400 personal depression and suicide risk assessment records from the elderly at the outpatient department, Thasala hospital and 9 district health promotion hospitals are located in Thasala district, Nakhon Si Thammarat. The results show that: 1) the depression assessment for the elderly 9Q data achieved at Overall Accuracy 95.25%, F-measure 83.72% and Average AUC 0.94 with 7 questions, 2) the elderly depression assessment for Thai Geriatric Depression Scale (TGDS) data achieved at Overall Accuracy 93.50%, F-measure 72.27% and average AUC 0.95 with only 17 questions, and 3) the suicide risk assessment for the elderly data achieved Overall Accuracy at 97.50%, F-measure 80.00% and Average AUC 0.87 with 9 questions. The results indicated that our method is highly significant in the prediction the depression and suicide risk in elderly.