Falling is an important factor that threatens people’s quality of life. With the aging of population all over the world, falling has become a problem that the society must pay attention to. If we can detect the symptoms of elderly falling, we can prevent them from falling. In order to predict falls more accurately, it is necessary to define and quantify the various factors influencing falls. In this thesis, we collected information through a questionnaire and thus defined the easy-to-fall-type, then we conducted experiments using a pressure distribution sensor mat to measure the walking data when the subject start walking, and finally analyzed the data through deep learning and performed validation.