A Fuzzy-based Neurodiagnostic System for Early Detection of Parkinson’s Disease using Wearable Sensor Data

Parkinson's Disease (PD) is a neurodegenerative disorder that necessitates early diagnosis to facilitate effective intervention. This paper suggests a Fuzzy-Based Neurodiagnostic System that utilizes the UCI Parkinson's Telemonitoring Dataset to classify early PD cases by integrating fuzzy logic and a deep neural network. 22 biomedical voice features, such as average fundamental frequency, jitter, shimmer, noise-to-harmonics ratio (NHR), and non-linear dynamical measures such as DFA and PPE, are included in the dataset. These features are captured from sustained phonation recordings. The interpretability of vocal irregularities is improved by the fuzzy logic module, which assigns fuzzy membership values (low, medium, high) based on clinical thresholds while processing three important uncertainty-prone features—jitter, shimmer, and NHR. These fuzzified outputs, in conjunction with the normalized raw features, are fed into a four-layer fully connected neural network (FCNN) that utilizes ReLU and Tanh activations. The first hidden layer consists of 64 neurons, while the subsequent layers contain 32 and 16 neurons, respectively. For binary classification (PD or non-PD), the output layer implements a Sigmoid activation. The model is trained using binary cross-entropy loss and 10-fold cross-validation, as well as the Adam optimiser. It outperformed conventional classifiers, including SVM and Decision Trees, with 94.8% accuracy, 92.3% precision, and 96.1% recall in this study. This hybrid soft computing approach effectively models uncertainty in voice features and extracts non-linear patterns, making it a promising diagnostic tool for real-time, noninvasive Parkinson's screening using ubiquitous sensor data.