Abstract / Summary
Abstract Parkinson's disease (PD) is a progressive disease that impacts movement and other functions. Voice disorders can present early in the disease process and voice analysis is an appealing non-invasive means of detection and monitoring of PD. The aim of the present review is to explore the application of machine learning (ML) and deep learning (DL) techniques to the diagnosis and prognosis of PD using voice signals. We applied the PRISMA 2020 guidelines to search ScienceDirect, IEEE Xplore, and ACM Digital Library for studies from the last eight years (2017–2025) with specific set criteria. A large number of public datasets, such as UCI Parkinson's data set and PC-GITA corpus, are included in this review. It briefly reviews some of the important acoustic characteristics, including jitter, shimmer, harmonic-to-noise ratio (HNR), mel-frequency cepstral coefficients (MFCCs), nonlinear dynamic measures, spectrogram features, and wavelet features. The review also covers the selection of features, reduction of data, and optimization of models. It provides a comparison of traditional ML models such as Support Vector Machine (SVM), Random Forest (RF), k-Nearest Neighbors (k-NN), Logistic Regression with advanced DL models including Convolutional Neural Networks (CNNs), recurrent neural networks, transformers and hybrid models that incorporate handcrafted and learned features. Last but not least, it shows how explainable artificial intelligence (XAI) can facilitate easier clinical interpretation of results. The results of most studies indicate over an 80% accuracy for PD detection using voice, indicating that the methods work. At the same time, there are issues including imbalanced data sets, scarcity of long-term studies, variations in study methodology, and a lack of generalizability between the data sets. The review proposes a standard and consistent approach along with several areas that could be explored in the future, such as integrating multiple types of data, conducting long-term studies, implementing federated learning, and building privacy-preserving AI systems for real-world clinical applications.