According to the World Health Organization (WHO), Parkinson's Disease (PD) is the second most common neurodegenerative condition that can cause tremors and other motor and non motor related symptoms. Medication and deep brain stimulation (DBS) are often used to treat tremor;​ however, medication is not always effective and has adverse effects, and DBS is invasive and carries a significant risk of complications. Wearable tremor suppression devices (WTSDs) have been proposed as a possible alternative, but their effectiveness is limited by the tremor models they use, which introduce a phase delay that decreases the performance of the devices. Additionally, the availability of tremor datasets is limited, which prevents the rapid advancement of these devices.

To address the challenges facing the WTSDs, PD tremor data were collected at the Wearable Biomechatronics Laboratory (WearMe Lab) to develop methods and data-driven models to improve the performance of WTSDs in managing tremor, and potentially to be integrated with the wearable tremor suppression glove that is being developed at the WearMe Lab. A predictive model was introduced and showed improved motion estimation with an average estimation accuracy of 99.2%. The model was also able to predict motion with multiple steps ahead, negating the phase delay introduced by previous models and achieving prediction accuracies of 97%, 94%, 92%, and 90% for predicting voluntary motion 10, 20, 50, and 100 steps ahead, respectively. Tremor and task classification models were also developed, with mean classification accuracies of 91.2% and 91.1%, respectively. These models can be used to fine-tune the parameters of existing estimators based on the type of tremor and task, increasing their suppression capabilities. To address the absence of a mathematical model for generating tremor data and limited access to existing PD tremor datasets, an open-source generative model was developed to produce data with similar characteristics, distribution, and patterns to real data. The reliability of the generated data was evaluated using four different methods, showing that the generative model can produce data with similar distribution, patterns, and characteristics to real data. The development of data-driven models and methods to improve the performance of wearable tremor suppression devices for Parkinson's disease can potentially offer a noninvasive and effective alternative to medication and deep brain stimulation. The proposed predictive model, classification model, and the open-source generative model provide a promising framework for the advancement of wearable technology for tremor suppression, potentially leading to a significant improvement in the quality of life for individuals with Parkinson's disease.