K-means clustering and LDA to identify changes in gait using a wearable device

Harvard University, ES155-Biological Signal Processing, Fall 2017

When walking with a load, such as a backpack, there may be changes in walking form that can cause poor posture and abnormal muscle stress. Using the Empatica E4 wrist band, blood volume pulse, acceleration, skin conductivity, and skin temperature were measured and analyzed to determine if a wristband can detect changes in gait due to loading. Subjects walked for 5 minutes on a treadmill, then 5 minutes on a treadmill while carrying a heavy backpack (40 lbs). Due to the constant treadmill speed, it was found that acceleration did not vary with time, but blood volume pulse increased through the duration of the trials. Using a peakfinding algorithm, the data was segmented into discrete steps, aligned, and averaged. LDA and K-means were used to differentiate normal walking and walking with a load using blood volume pulse and acceleration. Combing the additional data from other measurements, LDA was performed to accurately determine (n=2, 97% accuracy) when the subject was walking with or without a load. This could be done using multiple variables while using just acceleration or BVP were unable to differentiate the two scenarios.