Analysis and Classification of Tremors in a Parkinson's Disease Simulator Using Machine Learning

Hand tremor is a symptom typically observed in patients with Parkinson's disease (PD). However, repetitive hand movements in healthy patients or non-PD conditions can also be confused with this symptom. In that sense, this work develops a systematic analysis for differentiating the types of tremors in reference using Machine Learning techniques and a tremor simulator mechanism equipped with inertial sensors that will provide the necessary dataset for such analysis. According to scientific literature, this mechanism is based on frequency analysis with a principal component of about 5 Hz. The results show that the best classification model is the random forest with favorable metrics such as an accuracy of 98.66% and an F1 score of 98.66 %. This will allow a classification of the nature of tremors for subsequent application in diagnosing PD, reducing complexity in the clinical analysis through data collection with inertial sensors and applying an optimized algorithm. In addition, it means a step forward in automating clinical procedures to benefit patients with this type of disease with symptomatological particularities' such as hand tremors.