An Efficient Parkinson’s Disease Detection System Using Transfer Learning, Voting Classifier, and WebBased Implementation

Abstract

PD is a degenerative neurological illness that causes tremors, stiffness, and bradykinesia. Early and precise diagnosis is crucial for successful therapy and better patient quality of life. Diagnostic methods including clinical rating scales and neuroimaging are subjective, expensive, and inaccessible. This research offers an upgraded deep transfer learning-based Parkinson's disease diagnosis system employing handwritten spiral pictures to overcome these problems. A robust feature extraction technique using pretrained deep learning models like VGG19, ResNet50, and InceptionV3 captures complicated patterns from input photos. A genetic algorithm-based improved feature selection system finds the most important characteristics, eliminating redundancy and boosting classification performance. K-Nearest Neighbors (KNN), Enhanced KNN, Support Vector Machine (SVM), and Random Forest classifiers use the given characteristics. An ensemble Voting Classifier combines model capabilities to increase prediction performance. Experimental findings show that the Random Forest model outperforms state-of-the-art approaches with 100% accuracy, precision, recall, and F1-score. The system also uses a Flask web framework and SQLite database for safe signup and login and real-time illness prediction from spiral photos. The suggested early Parkinson's disease detection method is cost-effective, accurate, and accessible for real-world healthcare applications and remote diagnostic assistance systems.