Abstract: The present disclosure processes a user's medical data for each year to be input to a machine learning device for predicting dementia, and a data set composed of optimal features is constructed. The optimal features include at least information on the user's disease history, and the user's medical information for each year in the last 7 years or less. Precise prediction and diagnosis of dementia may be made by constructing the optimal features identified through experiments in the user's medical data for each year. Since the experimental results show that the prediction results of observing a disease history of 7 years or less may be the best, rather than observing medical information for a long period of time, the appropriate criteria may be suggested for predicting dementia.