Deep Transfer Learning Based Parkinson's Disease Detection Using Optimized Feature Selection

Abstract

Parkinson's Disease (PD) is a chronic and progressive neurological disorder affecting motor functions and quality of life. Early and accurate detection is critical for effective treatment planning. This paper proposes a deep transfer learning-based PD detection system using optimized feature selection from spiral handwriting images. Pre-trained CNNs — VGG19, InceptionV3, and ResNet50 — are used as feature extractors on the NEWHANDPD dataset. The extracted deep features are optimized using a Genetic Algorithm (GA) to select the most discriminative subset. The selected features are then classified using a K-Nearest Neighbor (KNN) classifier. Experimental results demonstrate that the proposed GA-optimized KNN model achieves an accuracy of 96–98%, significantly outperforming baseline SVM and standard KNN classifiers. The proposed approach provides an efficient, non-invasive, and cost-effective tool for early Parkinson's Disease screening using handwriting analysis.