Early Parkinson’s Disease Detection Through Handwriting Dynamics Analysis with Encoder Tuning Decoder Attention Network with Kookaburra Optimization Algorithm

Early Detection of Parkinson's Disease (EPDD) using Handwriting Dynamics Analysis (HDA) is a technique of handwriting pattern analysis (such as tremors, pressure, and speed) to detect motor control problems, supporting early diagnosis and intervention. Early diagnosis is critical for the effective treatment of PD and the enhancement of the patient's quality of life. Conventional deep learning cannot understand handwriting's temporal and spatial characteristics, compromising the accuracy of PD diagnosis. To overcome this challenge, this research suggests an Encoder Tuning Decoder Attention Network with Kookaburra Optimization Algorithm (ETDANet+KOA) to improve EPDD via HDA. The input data are obtained from the PaHaW dataset and are initially preprocessed by the Iterative Robust Peak-Aware Guided Filter (IRPAGF) method. Upon preprocessing, the feature extraction task is conducted using the Prompt-Tuned Multi-Task Taxonomic Transformer (PTM-TTT). Classification based on Encoder Tuning Decoder Attention Network (ETDANet) and subsequent optimization via Kookaburra Optimization Algorithm (KOA). The performance of the proposed ETDANet+KOA model is evaluated on the PaHaW dataset with a high accuracy rate of 99.9% and specificity of 99.8%. The proposed model is implemented based on the Python programming language. The result of the suggested ETDANet+KOA model illustrates superior accuracy and specificity in identifying early PD using HDA. The model successfully depicts temporal and spatial patterns of handwriting, enhancing performance in classification. The model demonstrates excellent reliability and robustness, reflecting its early detection and enhanced patient management possibilities.