Prediction of Grade, Gender, and Academic Performance of Children and Teenagers from Handwriting Using the Sigma-Lognormal Model

Adrian Iste, Kazuki Nishizawa, Chisa Tanaka, Andrew Vargo, Anna Scius-Bertrand, Andreas Fischer, Koichi Kise · 2026

Read Prediction of Grade, Gender, and Academic Performance of Children and Teenagers from Handwriting Using the Sigma-Lognormal Model by Adrian Iste, Kazuki Nishizawa, Chisa Tanaka, Andrew Vargo, Anna Scius-Bertrand, Andreas Fischer, Koichi Kise (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-11519.txt.

Introduction

Iste and co-authors ask what children's handwriting reveals about grade, gender, and school performance. Digital pens capture fine timing and motion signals from the act of writing. Most past work used clinical or adult samples, leaving child development understudied. The data set covers Japanese students from elementary through junior high school. Three feature families face off: plain kinematic statistics, entropy measures, and Sigma-Lognormal parameters. Grade comes through most clearly, the team finds, with weaker signals for the other two traits.

Catalog record

CREATORAdrian Iste, Kazuki Nishizawa, Chisa Tanaka, Andrew Vargo, Anna Scius-Bertrand, Andreas Fischer, Koichi Kise
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE31940
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.11519

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Prediction of Grade, Gender, and Academic Performance of Children and Teenagers from Handwriting Using the Sigma-Lognormal Model Adrian Iste, Kazuki Nishizawa, Chisa Tanaka, Andrew Vargo, Anna Scius-Bertrand, Andreas Fischer, Koichi Kise Abstract Digital handwriting acquisition enables the capture of detailed temporal and kinematic signals reflecting the motor processes underlying writing behavior. While handwriting analysis has been extensively explored in clinical or adult populations, its potential for studying developmental and educational characteristics in children remains less investigated. In this work, we examine whether handwriting dynamics encode information related to student characteristics using a large-scale online dataset collected from Japanese students from elementary school to junior high school. We systematically compare three families of handwriting-derived features: basic statistical descriptors of kinematic signals, entropy-based measures of variability, and parameters obtained from the sigma-lognormal model. Although the dataset contains dense stroke-level recordings, features are aggregated at the student level to enable a controlled comparison between representations. These features are evaluated across three prediction tasks: grade prediction, gender classification, and academic performance classification, using Linear or Logistic Regression and Random Forest models under consistent experimental settings. The results show that handwriting dynamics contain measurable signals related to developmental stage and individual differences, especially for the grade prediction task. These findings highlight the potential of kinematic handwriting analysis and confirm that through their development, children’s handwriting evolves toward a lognormal motor organization. 1 Introduction Handwriting is a complex motor activity resulting from the interaction between cognitive, perceptual, and neuromotor processes [ 18 ] . With the widespread adoption of digitizing devices, online handwriting acquisition now enables the capture of fine-grained temporal and kinematic signals, including pen velocity, pressure, tilt, and spatial trajectories. These measurements provide access to the dynamic aspects of writing behavior and offer new opportunities to study motor execution beyond the visual appearance of written traces. Unlike offline handwritten images, online recordings preserve the temporal structure of strokes and provide insight into the underlying motor control processes that generate written movements. Beyond character recognition, online handwriting data have been extensively utilized for authentication and biometric verification, writer identification, and the evaluation of writing skill and motor proficiency. These data have also facilitated developmental and aging studies by capturing the evolution of motor execution across the lifespan, as well as neurological and clinical assessments of disorders affecting motor control. Among modeling approaches, the sigma-lognormal model [ 17 ] offers a principled neuromotor framework that describes handwriting movements as the superposition of lognormal components associated with motor impulses. This model has demonstrated relevance in clinical contexts, including the assessment of neuromotor impairments such as Alzheimer’s disease [ 8 ] . Despite these advances, several potentially informative tasks remain insufficiently explored, especially in younger populations. Most prior studies have focused on adult writers or specific pathological conditions. In contrast, handwriting during childhood reflects ongoing motor learning, cognitive maturation, and progressive automation of movement patterns. Because children’s handwriting is still under development, it exhibits higher variability, less stable motor coordination, and evolving control strategies. As a result, tasks such as age prediction, gender classification, or academic performance classification may be particularly challenging in school-aged populations. In this work, we address these issues using a large-scale online handwriting dataset collected with Wacom tablets from Japanese students aged 7 to 15 years. The dataset includes approximately fifty writers per age group and around 110 drills per student. Rather than relying on computer vision approaches based on static images, we focus exclusively on features derived from writing movements themselves. We compare three families of representations: basic descriptors of kinematic signals such as speed and acceleration, entropy-based measures capturing variability and complexity in writing dynamics, and parameters derived from the sigma-lognormal model, which represents handwriting movements as the superposition of lognormal velocity primitives reflecting underlying motor commands. Although the recordings are available at the stroke level, we consider aggregated student-level representations in order to establish a first systematic comparison across feature types under consistent experimental conditions. We evaluate these feature sets across three prediction tasks reflecting different aspects of individual variability: grade prediction, gender classification, and academic performance classification, using linear models and random forest algorithms. The contributions of this study are as follows : we provide an analysis of online handwriting dynamics in a school-aged population, a demographic that remains underrepresented in prior work. Also, we investigate whether motor-oriented handwriting representations can serve as complementary indicators of developmental stage and academic-related characteristics. Through this analysis, we aim to better understand the potential of handwriting dynamics as a tool for studying development and learning processes in educational contexts, especially for young students. 2 Related Work The prediction of age and gender based on handwriting has been explored by various researchers. Alaei et al. offer a comprehensive review of existing methodologies, emphasizing the diversity of approaches, which range from traditional statistical analyses to advanced deep learning frameworks for these tasks [ 4 ] . Handwriting analysis has been approached through both offline and online datasets. For age prediction from handwriting, existing studies primarily address multi-class classification problems, typically distinguishing between “children” and “adults”. To the best of our knowledge, no research has focused on predicting features for children under 11 years of age. For age prediction, offline handwriting has been widely used : Basavaraja et al. used an offline dataset to predict the age of writers across four classes, with ages ranging from 11 to 24 years old [ 5 ] . They also employed the IAM dataset, which includes ages from 25 to 56 years, and the KHATT dataset, covering ages from 16 to 56 years. AL-Qawasmeh et al . created an offline handwriting dataset called FSHS comprising approximately 2000 writers aged from 15 to 75 [ 3 ] . They extracted features from the samples using CNN and attempted to predict the age and gender of the writers. For grade prediction, they proposed classifying writers into two categories: “young adult writer” and “mature adult writer”. Shin et al. employed an online dataset to predict the age class of writers, distinguishing between “child” (ages 12 to 13) and “adult” (ages 19 to 59), with approximately 80 participants. Features such as pen pressure, x and y coordinates of the pen, and pen angle were utilized [ 20 ] . Nunez et al. extracted entropy features from an online database (IRONOFF) consisting of approximately 800 participants aged 11 to 77, divided into six different classes. Nevertheless, since the class were unbalanced, they artificially reduced the number of writers to 144 [ 12 ] . For gender classification, most studies have been made using offline handwriting datasets : Maken et al. utilized the Kaggle offline dataset from the ICDAR 2013 gender prediction competition, which includes approximately 280 writers, each contributing two samples [ 14 ] . Additionally, Bi et al. [ 6 ] and Tan et al. [ 22 ] also employed this dataset for gender prediction, also using the RDF dataset, in Chinese, which includes about 11,000 writers, each providing one sample. Other researchers used the offline QUWI dataset, which includes 1000 writers and four samples per writer, written in Arabic and English [ 16 , 10 , 2 , 11 ] . Few studies have been made on gender classification using online handwriting

Read the complete text