Elderly life is full of hazards. The fall is one of the problems for elderly people. In this thesis, we intend to develop a fall risk prediction system on a smartphone which has an accelerometer. First, the system will record the user’s gait data sets and analyze the stability and the symmetry from the data sets. This thesis uses a high-level fuzzy Petri net (HLFPN) to identify human actions including normal action, exercising and fall. In total, there are 90 times of trials, and 62 times of successful prediction done by the system. We intend to decrease the probability to fall down for elderly people and the people who use the prostheses, and provide a safer environment for them.