Early Detection of Parkinson’s Disease using Wearable Sensor-based Gait analysis with Logistic Regression

In recent days, the early detection of Parkinson’s Disease (PD) from essential tremor (ET) is clinically difficult due to overlapping motor symptoms. The traditional models used wearable sensor-based gait analysis for structured test like Timed Up and Go (TUG) but struggled to control the clinical environments which limited the utility for continuous and real-world monitoring. Therefore, this research proposed an improved framework Wearable Sensor-based Gait analysis with Logistic Regression (WSGLR) for early detection of PD. The proposed WSGLR utilizes the sensor-based gait analysis to everyday settings with a reduced set of wearable Inertial Measurement Units (IMUs) placed on wrist and lower back. Then, input signals from sensors are preprocessed with Butterworth band-pass filter to avoid high-frequency noise and low frequency drift. Consequently, Bidirectional Long-Short Term Memory (BiLSTM) is employed for capturing temporal dependencies. Finally, a Logistic Regression (LR) model is introduced for early detection of PD from BiLSTM feature vectors. From the results, the proposed WSGLR attained better results in terms of accuracy 89% and Area Under Curve (AUC) of 0.952 respectively.