A Comprehensive Survey on Data Mining–Driven Approaches for Parkinson’s Disease Diagnosis and Analysis

Abstract

Parkinson Disease (PD) is a neurodegenerative progressive disorder that denotes motor deficits like tremor, rigidity, and bradykinesia and non-motor deficits like impaired speech and cognitive deterioration. Data mining approaches have attracted considerable interest in the recent years towards early diagnosis, severity, and progression monitoring of PD in the light of their capability to identify discriminative patterns in high-dimensional and heterogeneous biomedical data. This survey is a systematic review of data mining-based methods used to analyse PD among various data modalities, and these are: clinical records, voice signals, handwriting dynamics, gait parameters, neuroimaging data. SVM methods, random forests, k-nearest neighbors and neural networks are the most commonly used forms of supervised learning, which have been used to classify PD with impressive diagnostic accuracy on benchmark datasets. The methods of clustering and association rule mining that have been applied unsupervised have been used to discover latent symptom patterns, disease subtypes. Dimensionality reduction techniques (e.g. Principal Component Analysis) and feature selection techniques (e.g. mutual information and evolutionary optimization) are important to improve the generalization and interpretation of the models. Recent works are growing to combine deep learning with data mining pipelines to create feature extraction and model intricate temporal dependencies automation. Though such developments are made, there are issues with data imbalance, interpretability, inadequate longitudinal data, and clinical generalizability. This survey identifies the up-to-date trends, comparative, and open research issues, which will serve as a unified source of future research based on data mining on PD.

KEYWORDS

Parkinson’s disease diagnosis, medical data mining techniques, machine learning classification models, feature selection and dimensionality reduction, voice and gait signal analysis, biomedical pattern recognition systems, intelligent healthcare decision support, neurodegenerative disease prediction models.