Gender Estimation via Handwriting Dynamics: A Statistical and Machine Learning Approach

Abstract

Gender estimation from handwriting has long been of interest in forensic document examination. However, most prior studies rely on static handwriting features, with limited exploration of dynamic writing behaviour. The advent of advanced technologies in recording handwriting made it possible to study the dynamic nature of the handwriting. The present study investigates gender-related differences in dynamic handwriting features derived from digitally captured natural style-handwriting, integrating statistical analysis with machine learning classification. Handwriting samples were collected from 200 participants (100 males, 100 females) using a pen-enabled digital tablet across three trials. Five primary dynamic features: Time-stamped X and Y coordinate data, pressure, azimuth, altitude and time, were recorded and used to derive 21 kinematic, spatial, pressure-based, and temporal features. The statistical analysis using Mann-Whitney U test, indicated significant gender-related differences in several dynamic features, including temporal parameters (pen-up duration, pen-down duration, total duration), pressure-related measures, spatial (handwriting width and height) and kinematic parameters like velocity and acceleration. Features demonstrating statistical significance were subsequently employed as inputs to supervised machine learning models, including Random Forest, Support Vector Machine, and Gradient Boosting classifiers. The Random Forest achieved the highest classification accuracy (86.7%) followed by Support Vector Machine (82.7%) and Gradient Boosting (80%). The findings demonstrate the potential of dynamic handwriting features in forensic document examination for gender estimation and preliminary profiling of writers