Document Type : Research Paper

Authors

• Mohammad Amini ¹
• Alireza Tavakoli Targhi ²
• Mehdi Hosseinzadeh ^{3, 4}
• Faezeh Farivar ⁵
• Reza Bidaki ^{6, 7}

¹ Department of Computer Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran

² Department of Computer Science, Shahid Beheshti University, Tehran, Iran

³ Health Management and Economics Research Center, Iran University of Medical Sciences, Tehran, Iran

⁴ Computer Science, University of Human Development, Sulaymaniyah, Iraq

⁵ Department of Mechatronics Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran

⁶ Research Center of Addiction and Behavioral Sciences, Shahid Sadoughi University of Medical Sciences, Yazd, Iran

⁷ Diabetes Research Center, Shahid Sadoughi University of Medical Sciences, Yazd, Iran

Abstract

Developmental disorders are regularly observed in developmental writing skills (developmental dysgraphia) with considerable concerns. Physicians do their diagnosis on the basis of the juvenile’s written products as well as the attitudes and feedback taken from their teachers. This is a very laborious process and yet subjective in nature. Consequently, many juveniles suffering from this defect, particularly those with lower levels of the disorder remain undiagnosed. The aim of the present work was to find a new method for the automatic identification of dysgraphia even at minute levels. Utilizing the most sensitive pen tablet available to gather the desired dataset, we could extract all the considered datasets, i.e. temporal, spatial, kinematic, and pressure parameters with the greatest possible accuracy. On the whole, 102 students (both male and female) from the second, third, and fourth grades of primary schools were participated in the data collection phase by being asked to write a short paragraph. 51 students, in an age range of eight to ten years, were participated in each group, i.e. dysgraphic and non-dysgraphic. Next, a huge set of features (more than two thousand features) was extracted in the preprocessing phase. In the feature selection phase, we eventually ended up with sixteen features that proved to be the most effective in diagnosing dysgraphia. To distinguish between the dysgraphic and non-dysgraphic students, three different types of classifiers, i.e. random forests, AdaBoost classifiers, and support vector machines (SVM) were considered and compared. For the prediction of dysgraphia based on incessant handwriting features, the SVM was revealed to be the best model with a classification performance accuracy of 93.65%. Our work exhibited that online handwriting features including time, jerk, and altitude/azimuth may be utilized to automatically reveal dysgraphia in juveniles with this writing disorder.

Keywords

• Diagnosis of Dysgraphia
• Online Handwriting Features
• Handwriting Analysis
• Machine Learning
• Classification