With the speedy aging population, weakening family endowment and unsound social endowment services, old-age care faces many challenges in China. In order to provide suitable living arrangement suggestions to the elderly and accurate decision-supporting to the departments geared toward elderly care, we analyze nearly 20,000 old people's data in CHARLS questionnaire, trying to find main factors affecting the elderly living preference. Besides, we also attempt to predict living preferences for old people by using the characteristics data of the elderly and improve feature selection algorithm on imbalanced data based on random forests. Experimental results indicate that the elderly living preference can be predicted well by using the characteristics data of the elderly. Importantly, this method is potential to provide reference on accuracy decision-making for the departments geared toward elderly care.