Knowledge Discovery in a Community Data Set: Malnutrition among the Elderly

OBJECTIVES: The purpose of this study was to design a prediction model that explains the characteristics of elderly adults at risk of malnutrition. METHODS: Data were obtained from a large data set, 2008 Korean Elderly Survey, in which the data of 15,146 subjects were entered. With nutritional status a target variable, the input variables included the demographic and socioeconomic status of participants. The data were analyzed by using the SPSS Clementine 12.0 program's feature selection node to select meaningful variables. RESULTS: Among the C5.0, C&R Tree, QUEST, and CHAID models, the highest predictability was reported by C&R Tree with the accuracy rate of 77.1%. The presence of more than two comorbidities, living alone status, having severe difficulty in daily activities, and lower perceived economic status were identified as risk factors of malnutrition in elderly. CONCLUSIONS: A reliable decision support model was designed to provide accurate information regarding the characteristics of elderly individuals with malnutrition. The findings demonstrated the good feasibility of data mining when used for a large community data set and its value in assisting health professionals and local decision makers to come up with effective strategies for achieving public health goals.