Detection of the source of Parkinson treatment inefficiency using machine learning based on tremor signals

Parkinson’s disease (PD) is a progressive neurological illness that primarily impairs movement. Deep brain stimulation (DBS) therapy and medication are frequently used in conjunction for the treatment of PD. However, achieving the best therapeutic results is difficult due to patient variability in treatment response and the complex interaction between medication and DBS efficacy. In order, to clarify the variables affecting treatment response, this study thoroughly examines the ON/OFF states of DBS, medication, and their combined effects in PD patients. 16 patients were concerned under 70 years with PD. In this paper, we use eight features derived from tremor signals, and ten classification models to analyze the treatment effects. The classification is between three classes: DBS OFF, medication OFF, and medication & DBS OFF. After applying the classifier on the extracted features without augmentation, we got the best classification for LDA with an accuracy of 72%, after doing a signal augmentation, we got the best classification for XGBOOST, DT, and Random Forests with an accuracy of 98% and after doing a features augmentation, we got the best classification for SVM, KNN, DT, Random Forests, and XGBOOST with an accuracy of 100%. These findings highlight the potential of augmented features and advanced classification models in optimizing PD treatment strategies.