Realtime Gait Phase Prediction and Early Detection of Gait Abnormalities

A gait cycle of walking can be viewed as a sequence of phases. Degree of movement of muscles involved in gait from one step to another can be analyzed to understand various gait patterns. In this research, gyroscope sensors have been used in experimental setup for recording gait phase data. The present study aims to propose a prototype for real-time prediction of next gait phase and classification of gait phase as normal or abnormal with a portable electronic circuit. Prediction model configured with convolution-based attention paired with Long Short-Term Memory (LSTM) showed very good performance metrices. Classification model configured with Artificial Neural Network (ANN) has been pretrained using preexisting dataset and transfer learning has been used for classification of predicted gait phase. Various machine learning techniques like exploratory data analysis, feature engineering, feature selection and data normalization has been used to improve prediction. This work successfully proposes a noble technique for early detection of gait abnormalities.