Affiliations: [a]
Department of Computer Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran
| [b]
Department of Computer Science, Shahid Beheshti University, Tehran, Iran
| [c]
Health Management and Economics Research Center, Iran University of Medical Sciences, Tehran, Iran
| [d] Computer Science, University of Human Development, Sulaymaniyah, Iraq
| [e]
Department of Mechatronics Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran
| [f]
Research Center of Addiction and Behavioral Sciences, Shahid Sadoughi University of Medical Sciences, Yazd, Iran
| [g]
Diabetes Research Center, Shahid Sadoughi University of Medical Sciences, Yazd, Iran
Correspondence:
[*]
Corresponding author. Alireza Tavakoli Targhi, Department of Computer Science, Shahid Beheshti University, Tehran, Iran. E-mail: a_tavakoli@sbu.ac.ir.
Abstract: Handwriting problems, also known as dysgraphia, are defined as a disorder or difficulty in producing written language associated with writing mechanics. The occurrence of handwriting problems among elementary students varies from 10 to 34%. With negative impacts on educational performance, handwriting problems cause low self-confidence and disappointment in the students. In this research, a pen-tablet was employed to sample children’s handwriting, which revealed online features of handwriting such as kinematic and temporal features as well as wrist and hand angles and pen pressure on the surface. This digitizer could also extract the online handwriting features when the pen was not in contact with the surface. Such features are called in-air features. The purpose of this study was to propose a method for diagnosing dysgraphia along with an evaluation of the impact of in-air features on the diagnosis of this disorder. A rich dataset (OHF-1) of online handwriting features of dysgraphic and non-dysgraphic students was prepared. After the extraction of a huge set of features and choosing a feature selection method, three machine learning methods, i.e. SVM, Random Forest and AdaBoost were compared and with the SVM method, an accuracy of 85.7% in diagnosing dysgraphia was achieved, when both in-air and on-surface features were included. However, while using purely in-air data or merely on-surface features, accuracies of 80.9% and 71.4% were achieved, respectively. Our findings showed that in-air features had a significant amount of information related to the diagnosis of dysgraphia. Consequently, they might serve as a significant part of the dysgraphia diagnosis.
Keywords: Handwriting Analysis, Identification of Dysgraphia, In-Air Analysis, Machine Learning, Online Handwriting Features
