Foot Strike Detection in Pathological Gait Using Random Dilated Shapelet Transform Classifier

Abstract

Accurate Automated foot strike detection for pathological gait is a known research problem. Sequence to Sequence LSTM has been used in research for foot strike event detection in pathological gait for heel strike patients. Kinematics produced from three-dimensional gait analysis or sensor data is used as input for event detection. Deep learning models require a larger dataset and are complex to train. Shapelets are short subsequences of time series that better represent a class. Shapelet based algorithms are fast to train, interpretable and can be trained for a dataset of modest size. We propose to use Random Dilated Shapelets transform combined with Rule based Detection Algorithm in order to detect foot strike for patients with heel strike foot contact pattern. The resultant model is interpretable, efficientand  has an accuracy of more than 90% .