The purpose of this research is to detect ‘off phenomenon’ that occurs when a medicine for Parkinson's disease loses its effectiveness. We focus on the periodicity of walking in patients with Parkinson's disease. A patient is requested to attach a simple device equipped with an accelerometer, and acceleration time series during walking will be collected. Our idea is to employ LSTM to detect walking abnormalities. To examine the effectiveness, we conducted some experiments. Firstly, two Parkinson's disease patients in on-state were asked to perform two 60-seconds walks. Secondly, a healthy person performed pseudo-off-state abnormal gait by attaching rubber restraints to his feet. Finally, a model was generated by the time series of acceleration in off-state patients' walk. The model was then used to calculate Mahalanobis distance to examine anomalies by comparing the model to the pseudo off-state walk data. As a result, the distance exhibited significantly high values at the pseudo off-state walk phases, which suggests that the proposed simple method will detect off-state walking by actual Parkinson’s disease patients.