Scholarlink

In recent years, eye-tracking technology has emerged as a promising non-invasive tool for assessing cognitive and reading difficulties, particularly dyslexia. By analyzing subtle differences in eye movement patterns, such as fixations, distortion, and saccades, this approach offers new possibilities for early detection and intervention. This study focuses on utilizing eye-tracking data from 84 participants screened using the Dyslexia Screener for Adults to classify dyslexic, non-dyslexic, and at-risk individuals. Critical eye movement features, including the X and Y coordinates of both eyes, fixation durations, and saccade patterns, were extracted and processed. Following event detection, windowing, and feature extraction, the dataset was expanded to 9400 training samples and 2351 testing samples for robust model training and evaluation. Series of experiments were conducted with and without feature selection (such as univariate f-classif, mutual info, RFE, Lasso, tree-based methods) to assess its impact on classifier performance. Machine learning classifiers, including Decision Trees (DT), Support Vector Machines (SVM), k-Nearest Neighbors (KNN), and ensemble methods, were applied. Results showed that ensemble methods and KNN achieved classification accuracies exceeding 80%, effectively differentiating between dyslexic, non-dyslexic, and at-risk subjects. Experimental results demonstrate that after employing various feature selection techniques, there is improvement in the accuracy of classifiers. Results suggest that using feature set of top 20 features further optimized model performance by selecting the most informative features. Among all classifiers, Ensemble Subspace Discriminant classifier, an ensemble model attained the highest testing accuracy of 87.91%, with a precision of 88.04% and recall of 87.91%. Though models demonstrated strong classification performance, the study acknowledges limitations due to the small dataset size, which may impact the generalizability of findings to larger populations. Nonetheless, the results emphasize and highlight the potential of advanced machine learning, especially ensemble learning models combined eye-tracking technology in early dyslexia detection, paving way for future interventions in large and more diverse samples.

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

Enhancing Dyslexia Classification Using Feature Selection and Ensemble Learning Models on Eye-Tracking Data

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Tabassum Gull Jan,
  • Sajad Mohammad Khan,
  • Sajid Yousuf Bhat

摘要

In recent years, eye-tracking technology has emerged as a promising non-invasive tool for assessing cognitive and reading difficulties, particularly dyslexia. By analyzing subtle differences in eye movement patterns, such as fixations, distortion, and saccades, this approach offers new possibilities for early detection and intervention. This study focuses on utilizing eye-tracking data from 84 participants screened using the Dyslexia Screener for Adults to classify dyslexic, non-dyslexic, and at-risk individuals. Critical eye movement features, including the X and Y coordinates of both eyes, fixation durations, and saccade patterns, were extracted and processed. Following event detection, windowing, and feature extraction, the dataset was expanded to 9400 training samples and 2351 testing samples for robust model training and evaluation. Series of experiments were conducted with and without feature selection (such as univariate f-classif, mutual info, RFE, Lasso, tree-based methods) to assess its impact on classifier performance. Machine learning classifiers, including Decision Trees (DT), Support Vector Machines (SVM), k-Nearest Neighbors (KNN), and ensemble methods, were applied. Results showed that ensemble methods and KNN achieved classification accuracies exceeding 80%, effectively differentiating between dyslexic, non-dyslexic, and at-risk subjects. Experimental results demonstrate that after employing various feature selection techniques, there is improvement in the accuracy of classifiers. Results suggest that using feature set of top 20 features further optimized model performance by selecting the most informative features. Among all classifiers, Ensemble Subspace Discriminant classifier, an ensemble model attained the highest testing accuracy of 87.91%, with a precision of 88.04% and recall of 87.91%. Though models demonstrated strong classification performance, the study acknowledges limitations due to the small dataset size, which may impact the generalizability of findings to larger populations. Nonetheless, the results emphasize and highlight the potential of advanced machine learning, especially ensemble learning models combined eye-tracking technology in early dyslexia detection, paving way for future interventions in large and more diverse samples.