The importance of integral time length windows for the classification of activities of daily living based on Machine Learning techniques

Pathological tremor, a prevalent movement disorder seen in essential tremor (ET) and Parkinson’s disease (PD) patients, is the most common tremor disorder impacting the quality of life of those who suffer from it. This study proposes a method to classify daily life activities using a single wrist-worn IMU for tremor patients. The used dataset involves IMU recordings from the dominant arm during 11 tasks performed by ET and PD patients. Signal features were extracted from different sized windows and used to train Random Forest (RF) and Support Vector Machine (SVM) models, training 10 different models overall. Results shows that although larger window sizes, particularly the 10 seconds window, provided highest average F1-score, certain specific activities were better classified with shorter windows. This approach outperforms prior studies by achieving improved classification outcomes and opens a new line in continuous tremor monitoring. Future research could explore the combination of various window lengths to identify optimal window durations for further accuracy refinement.