Severity Classification of Freezing of Gait Using Machine-Learning Algorithms: A Hidden State Model Approach
E-health sensors and wearables play an important role in the detection and classification of many chronic diseases. A chronic disease requires active monitoring and its severity increases over time. Parkinson’s disease is one such chronic disease resulting in motor as well as non-motor im...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11087574/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | E-health sensors and wearables play an important role in the detection and classification of many chronic diseases. A chronic disease requires active monitoring and its severity increases over time. Parkinson’s disease is one such chronic disease resulting in motor as well as non-motor impairments. Freezing of gait (FOG) is one of the motor symptoms that could affect the quality of life of patients. Diagnosing the presence and severity of Parkinson’s disease and its symptoms can be done subjectively and objectively. However, subjective analysis might suffer from biases and variability as it involves human emotions and need objective method or measurable data to make decisions. So, in this regard, we conducted a feasibility study by adopting an objective approach for labeling the FOG severity levels and then classifying them using the data from various sensors. We use the Hidden Markov Model (HMM) for labeling the severity levels. We also trained an ensemble algorithm to model the prediction of severity levels marked by HMM model. We could obtain severity level classification accuracy between <inline-formula> <tex-math notation="LaTeX">$87-88\%$ </tex-math></inline-formula> using the XGBoost algorithm. Moreover, we also used model explainability approach to explain the predictions on severity levels using only objective methods, in this case sensor data. We found that features from EEG show a positive correlation with severe FOG events and gyroscope sensors show a negative correlation for predicting severe FOG events. |
|---|---|
| ISSN: | 2169-3536 |