Deep Learning-Based Classification of Handwriting Data: A Multi-Representation Approach For Alzheimer’s Disease

The study of handwriting behavior is becoming an increasingly important method for assessing neurodegenerative disorders. In this study, a feature set was created by recalculating 18 features (including temporal, kinematic, and geometric characteristics) defined in the literature from raw data in the DARWIN handwriting database. The resulting representations was initially evaluated using a 1-Dimensional Convolutional Neural Network (1D-CNN) to investigate the information provided by handwriting patterns for the detection of Alzheimer's disease. In the second stage, this feature set was converted into a two-dimensional image format, and the visual representations were further analyzed using the 2D Convolutional Neural Network (2D-CNN) model. The 1D-CNN model developed on numerical data achieved 94.29% accuracy, while the 2D-CNN model trained on two-dimensional visual representations achieved 91.43% accuracy. The findings indicate that numerical and visual representations derived from handwriting data contain significant discriminatory information for the classification of Alzheimer's disease and that both approaches exhibit similar and comparable levels of performance.