A Comparative Study of Machine Learning and Deep Learning Models for Early Detection of Parkinson's Disease Using Voice Features
DOI:
https://doi.org/10.61952/jlabw.v1i4.216Keywords:
Parkinson's Disease, Voice Signals, Machine Learning, CRISP-DM Methodology, Explainable AI (XAI), Feature Importance, PPE, RPDEAbstract
his study aims to present a detailed and reliable methodology for preparing and analyzing acoustic data for the early detection of Parkinson's Disease (PD). The research adopted the standard CRISP-DM framework to structure the workflow, starting from integrating a multi-source voice dataset, through essential preprocessing steps such as standardization and data splitting. To overcome the "black box" problem and enhance clinical trust, the research focused on applying Explainable AI (XAI) principles. The winning Random Forest (RF) model from the initial study was utilized to determine and assess the relative importance of 16 acoustic features. The analysis demonstrated that non-linear features, specifically PPE (Pitch Period Entropy) and RPDE (Recurrence Period Density Entropy), are the most influential factors in the diagnostic decision. This confirms that changes in voice signal complexity are powerful biomarkers for the disease. This work provides a clear, systematic, and interpreted roadmap, recommending prioritizing these features for developing effective and safe automated diagnostic systems.
References
Ahmadi, A., et al. (2022). Deep learning for detecting Parkinson’s disease from speech signals using MFCC features. Biomedical Signal Processing and Control, 72, 103289. https://doi.org/10.1016/j.bspc.2021.103289
Berry, M. J. A., & Linoff, G. S. (2004). Data Mining Techniques: For Marketing, Sales, and Customer Relationship Management (2nd ed.). Wiley.
Kim, J., et al. (2024). Explainable machine learning model for Parkinson’s disease diagnosis from voice features. Computers in Biology and Medicine, 169, 107675. https://doi.org/10.1016/j.compbiomed.2024.107675
Lage, I., et al. (2019). Interpretable machine learning for healthcare: A comparative study of explanations generated for clinical predictive models. Bioinformatics, 35(17), 2959–2967. https://doi.org/10.1093/bioinformatics/btz084
Little, M. A., et al. (2019). The role of nonlinear dynamics in diagnosing voice disorders from sustained phonation. Journal of Voice, 33(2), 195.e1–195.e10.
Singh, R., & Samaria, D. (2021). A systematic review on Parkinson’s disease detection using speech and machine learning techniques. Journal of King Saud University – Computer and Information Sciences, 33(10), 1185–1195. https://doi.org/10.1016/j.jksuci.2019.06.010
Tsanas, A., & Little, M. A. (2009). Accurate automated diagnosis of Parkinson’s disease from sustained vowels. In 2009 International Conference on Machine Learning and Applications (pp. 714–719). IEEE.
Tsanas, A., et al. (2020). Novel speech signal processing algorithms for robust detection of Parkinson’s disease. IEEE Transactions on Biomedical Engineering, 57(12), 2955–2966.
Zhao, H., et al. (2023). Integrating multi-source voice datasets with CRISP-DM methodology for Parkinson’s disease detection. Artificial Intelligence in Medicine, 140, 102635. https://doi.org/10.1016/j.artmed.2023.102635
Abdusamea Omer, Rabyah B. Ali, Ahmed Al-Siddiq Masoud Al-Dabbashi, & Ali Abdulhamid Ali Al-Halak. (2025). A Comparative Study of Machine Learning and Deep Learning Models for Early Detection of Parkinson’s Disease Using Voice Features. African Journal of Advanced Pure and Applied Sciences (AJAPAS), 4(4), 106–113.
