• Version
• Download 17
• File Size 411.30 KB
• File Count 1
• Create Date 28 January 2026
• Last Updated 28 January 2026

Early Detection of Parkinson’s Disease using CNN And RNN

Kishore Kumar Kalli
Department of ArtificialIntelligence and Data Science
Koneru Lakshmaiah Education Foundation
Vaddeswaram
2100080178ai.ds@gmail.com

Gadugolu Roshitha
Department of ArtificialIntelligence and Data Science
Koneru Lakshmaiah Education Foundation
Vaddeswaram
2100080198ai.ds@gmail.com

Dr. A. Srinivasa Rao
Department of ArtificialIntelligence and Data Science
Koneru Lakshmaiah Education Foundation
Vaddeswaram
asrinivasarao1@kluniversity.in

Parkinson's Disease (PD) presents a significant challenge due to its progressive nature and the array of motor and non-motor symptoms it entails. One of the early manifestations of PD often includes vocal impairments, making diagnosis systems that leverage vocal features crucial in early detection efforts. Our study proposes two distinct frameworks based on Convolutional Neural Networks (CNNs) aimed at classifying PD using sets of vocal features. The distinguishing factor between these frameworks lies in their approach to combining these feature sets. In the first framework, feature sets are amalgamated before inputting them into a 9-layered CNN. Conversely, the second framework adopts a parallel input layer strategy, wherein feature sets are fed directly into separate convolution layers, allowing for simultaneous extraction of deep features before merging them in a subsequent layer.

The effectiveness of our proposed models was evaluated using a dataset sourced from the UCI Machine Learning repository, with performance validation conducted through Leave-One-Person-Out Cross Validation (LOPO CV). Given the imbalanced class distribution within our data, we employed F-Measure and Matthews Correlation Coefficient metrics alongside accuracy for comprehensive assessment. Experimental results showcased promising outcomes, particularly with the second framework, which demonstrated superior performance. By leveraging parallel convolution layers, this framework efficiently captured deep features from each feature set, enhancing the classifiers' discriminative power. Notably, the extracted deep features not only effectively differentiated PD patients from healthy individuals but also significantly bolstered the overall classification efficacy.

The comparison between the two proposed frameworks highlights the efficacy of the parallel input layer strategy, particularly in capturing and leveraging deep features from distinct feature sets. These findings not only contribute to the advancement of PD detection systems but also emphasize the importance of innovative approaches in mitigating the challenges posed by imbalanced data distributions and heterogeneous symptom profiles within neurodegenerative diseases like PD. Experimental results showcased promising outcomes, particularly with the second framework, which demonstrated superior performance. By leveraging parallel convolution layers, this framework efficiently captured deep features from each feature set, enhancing the classifiers' discriminative power. Notably, the extracted deep features not only effectively differentiated PD patients from healthy individuals but also significantly bolstered the overall classification efficacy.

Keywords— convolutional neural networks ,deep learning ,health informatics, Parkinson’s Disease

classification, vocal features.