Footstrike pattern recognition using machine learning on tibial accelerometry

Recent reports have suggested that high loading rates, typically associated with a rearfoot strike (RFS) pattern, during running may be related to overuse injuries. With the recent increase in wearable sensors it is important to identify paradigms where footstrike pattern (FSP) can be detected from minimal data. Two machine learning (ML) classification techniques, Support Vector Machines (SVM) and Neural Networks (NN), were employed to identify running FSP using tibial acceleration (TA). SVM and NN models were trained and validated using this data set. These machine learning techniques were able to correctly identify FSP 86.1 to 96.4% of the time.