Digital handwriting metrics and temporal behaviors for predicting early Alzheimer’s onset

Aim: The aim of this paper was to explore stroke-level handwriting dynamics as early behavioral biomarkers for Alzheimer’s disease (AD). Methods: Stroke-level handwriting data were collected from 174 participants (89 probable AD or mild cognitive impairment; 85 cognitive healthy controls). Temporal, kinematic, and pressure features were extracted and aggregated. Classification performance was evaluated using Logistic Regression, Support Vector Machine (SVM), and Random Forest (RF) under five-fold stratified cross-validation. Random forest with SHapley Additive exPlanations (SHAP) was used to interpret feature contributions and temporal trends. Results: Participants with AD exhibited longer and more variable in-air times, slower stroke speed, and higher-pressure variability. SVM achieved the highest ROC-AUC (0.923), while Random Forest demonstrated robust and balanced performance (accuracy = 0.844) and identified key predictive features (mean and variability of in-air time and pressure). Temporal analysis revealed progressive motor hesitation across strokes. Conclusions: Stroke-level handwriting dynamics provide sensitive and interpretable biomarkers for early Alzheimer’s disease detection. Variability in in-air time and related temporal features effectively distinguish Alzheimer’s disease patients from cognitively healthy controls, reflecting underlying motor–cognitive coupling deficits. These findings highlight digital handwriting analysis as a scalable, non-invasive approach for early screening and monitoring, with potential to detect subtle impairments that may precede clinical symptoms.