Handling Class Imbalance in Forecasting Parkinson’s Disease Wearing-off with Fitness Tracker Dataset

Parkinson’s disease (PD) patients experience the “wearing-off phenomenon”, where their symptoms resurface before they can take the following medication. As time passes, the duration of the medicine’s efficacy reduces, leading to discomfort among PD patients. Therefore, patients and clinicians must meticulously observe and document symptom changes to administer appropriate treatment. Forecasting the PD wearing-off phenomenon is challenging due to the class imbalance from the difficulty documenting the phenomenon. This paper compares different approaches for handling class imbalance in forecasting the PD wearing-off phenomenon using the fitness tracker and smartwatch dataset (wearing-off dataset): oversampling, undersampling, and combining the two. Previous studies reported the potential use of commercially-worn fitness tracker datasets to predict and forecast wearing-off periods. However, some participants’ high false positives and negatives have been observed with the developed models [16, 17]. This paper compares different approaches to handling class imbalance in the wearing-off dataset. First, changes in the data collection process and tools were made during the data collection phase, as the nursing staff struggled with the data collection tool. Second, different existing oversampling and undersampling techniques were used to improve the ratio of wearing-off labels to non-wearing-off instances. Finally, adjustments to forecast probabilities were applied due to the resampling in the second step. The results showed that an XGBoost model trained with an oversampled dataset using the SMOTE algorithm increased the F1 and AUC-ROC curve scores by 7.592% and 0.754%, respectively, compared to the based XGBoost model. The models in this paper only used the last hour’s data to forecast the next hour’s wearing-off. The change in the amount of input data also improved the F1 and AUC-ROC curve scores compared with the models that used the current period or the previous day’s data [16, 17]. These improvements in forecasting wearing-off can help PD patients and clinicians monitor, record, and prepare for wearing-off periods, as we aim to lessen the false positives and negatives in the forecasting results.