Classifying Foot Strike Pattern using Tibia Mounted Accelerometer through Machine Learning

Running can lead to injuries due to high-impact loading on the feet, but changing foot strike (FS) may help some runners avoid this. Using tibia-mounted uni-axial accelerometer data from 58 runners, two machine-learning models (Neural Network (NN) and Short-Term Long memory (LSTM)) were developed to identify FS. Further optimization through Keras Tuner, leave two out, max voting, and random splitting, the NN achieved an average of 86% testing accuracy, and the LSTM model achieved 66%.