Scholarlink

<p>Developmental dysgraphia is a neurological disorder that impairs children’s writing abilities. Its heterogeneous symptoms and frequent co-occurrence with other disorders make accurate diagnosis challenging. In recent years, researchers have explored machine learning approaches for dysgraphia diagnosis using either offline or online handwriting data. However, most studies analyze these modalities separately, limiting the ability to capture the relationship between them. To address this gap, we propose the first multimodal machine learning approach that integrates both online and offline handwriting data. We created a novel multimodal dataset by transforming an existing online handwriting dataset into its offline counterpart, generating corresponding handwriting images. Our analysis focuses on different word types, including simple words, pseudowords, and difficult words. We trained SVM and XGBoost classifiers separately on online and offline features, then implemented multimodal feature fusion and soft-voted ensemble methods. Additionally, we introduce a novel ensemble with conditional feature fusion, which selectively integrates feature-level fusion when confidence scores fall below a threshold. Our proposed approach achieves an accuracy of 88.8%, outperforming single-modality SVMs by 12–14 percentage points, existing methods by 8–9 percentage points, and traditional multimodal approaches (soft-vote ensemble and feature fusion) by 3 and 5 percentage points, respectively. This methodology contributes to the preliminary screening of handwriting difficulties through multimodal analysis of word/pseudoword data, offering an efficient initial assessment tool that supports dysgraphia diagnosis. Our findings highlight the potential of multimodal learning in enhancing dysgraphia assessment, paving the way for more accessible and effective diagnostic tools.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multimodal Ensemble with Conditional Feature Fusion for Dysgraphia Diagnosis in Children from Handwriting Samples

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月24日

查看全文
  • Jayakanth Kunhoth,
  • Moutaz Saleh,
  • Somaya Al-Maadeed,
  • Younes Akbari

摘要

Developmental dysgraphia is a neurological disorder that impairs children’s writing abilities. Its heterogeneous symptoms and frequent co-occurrence with other disorders make accurate diagnosis challenging. In recent years, researchers have explored machine learning approaches for dysgraphia diagnosis using either offline or online handwriting data. However, most studies analyze these modalities separately, limiting the ability to capture the relationship between them. To address this gap, we propose the first multimodal machine learning approach that integrates both online and offline handwriting data. We created a novel multimodal dataset by transforming an existing online handwriting dataset into its offline counterpart, generating corresponding handwriting images. Our analysis focuses on different word types, including simple words, pseudowords, and difficult words. We trained SVM and XGBoost classifiers separately on online and offline features, then implemented multimodal feature fusion and soft-voted ensemble methods. Additionally, we introduce a novel ensemble with conditional feature fusion, which selectively integrates feature-level fusion when confidence scores fall below a threshold. Our proposed approach achieves an accuracy of 88.8%, outperforming single-modality SVMs by 12–14 percentage points, existing methods by 8–9 percentage points, and traditional multimodal approaches (soft-vote ensemble and feature fusion) by 3 and 5 percentage points, respectively. This methodology contributes to the preliminary screening of handwriting difficulties through multimodal analysis of word/pseudoword data, offering an efficient initial assessment tool that supports dysgraphia diagnosis. Our findings highlight the potential of multimodal learning in enhancing dysgraphia assessment, paving the way for more accessible and effective diagnostic tools.