Background

Dyslexia is a disorder characterized by difficulty in reading such as poor speech and sound recognition. They have less capability to relate letters and form words and exhibit poor reading comprehension. Eye-tracking methodologies play a major role in analyzing human cognitive processing. Dyslexia is not a visual impairment disorder but it's a difficulty in phonological processing and word decoding. These difficulties are reflected in their eye movement patterns during reading.

Objective

The disruptive eye movement helps us to use eye-tracking methodologies for identifying dyslexics.

Methods

In this paper, a small set of eye movement features have been proposed that contribute more to distinguish between dyslexics and non-dyslexics by machine learning models. Features related to eye movement events such as fixations and saccades are detected using statistical measures, dispersion threshold identification (I-DT) and velocity threshold identification (I-VT) algorithms. These features were further analyzed using various machine learning algorithms such as Particle Swarm Optimization (PSO) based SVM Hybrid Kernel (Hybrid SVM – PSO), Support Vector Machine (SVM), Random Forest classifier (RF), Logistic Regression (LR) and K-Nearest Neighbor (KNN) for classification of dyslexics and non-dyslexics.

Results

The accuracy achieved using the Hybrid SVM –PSO model is 95.6 %. The best set of features that gave high accuracy are average no of fixations, average fixation gaze duration, average saccadic movement duration, total number of saccadic movements, and average number of fixations.

Conclusion

It is observed that eye movement features detected using velocity-based algorithms performed better than those detected by dispersion-based algorithms and statistical measures.

中文翻译：

---

从注视和眼跳运动中发现诵读困难的预测模型。

背景

诵读困难症是一种以阅读困难为特征的疾病，例如不良的言语和声音识别。他们缺乏关联字母和形成单词的能力，并且阅读能力较差。眼动追踪方法在分析人类认知过程中起主要作用。阅读障碍不是视觉障碍，但在语音处理和单词解码方面是困难的。这些困难反映在阅读过程中他们的眼睛运动方式上。

目的

破坏性的眼球运动有助于我们使用眼动追踪方法来识别阅读困难。

方法

在本文中，已经提出了一小部分眼睛运动特征，这些特征通过机器学习模型对区分阅读障碍和非阅读障碍做出了更大贡献。使用统计测量，色散阈值识别（I-DT）和速度阈值识别（I-VT）算法来检测与眼睛运动事件有关的特征，例如注视和扫视。使用各种机器学习算法，例如基于粒子群优化（PSO）的SVM混合内核（Hybrid SVM – PSO），支持向量机（SVM），随机森林分类器（RF），对数回归（LR）和K -最近邻居（KNN），用于对阅读障碍和非阅读障碍进行分类。

结果

使用Hybrid SVM –PSO模型获得的精度为95.6％。能够提供高精度的最佳功能集是平均注视次数，平均注视凝视时间，平均扫视运动持续时间，扫视运动总数和平均注视次数。

结论

可以观察到，使用基于速度的算法检测到的眼睛运动特征要比基于色散的算法和统计测量所检测到的更好。