A Two Layer Hybrid Approach for Parkinson’s Disease Detection Optimized via Modified Metaheuristic Algorithm

Parkinson’s disease is one of the more frequent disorders affecting the nervous system, primarily affecting motor functioning, but followed by executive dysfunction and, in some cases, dementia. While no cure is available yet, early detection and treatment do have an immense impact on the quality and length of life. Since the symptoms are hard to detect in the early stages, applying machine learning can enhance the diagnostics. Long short-term memory neural networks are employed for their capability to utilize sequential data, with the added help from XGBoost on the last architectural level for higher efficacy of the model forming a collaborative hybrid model. Hyperparameters are crucial for the model’s accuracy and efficiency, so a modified hybridized metaheuristic algorithm is created and applied to the task. The results validate the model, as the presented approach achieved an accuracy exceeding 89% suggesting that the dual-layer hybrid approach introduced in this work has the potential to aid in the early detection of neurodegenerative conditions based on non-invasive collected data from shoe-mounted sensors.