Color Vision Deficiency Recognition Based on Eye-Tracking Metrics Using Machine Learning Approaches

Individuals with color vision deficiency (CVD) face many difficulties and limitations in their daily lives and professional activities, some of which may prove life-threatening. These negative factors necessitate the development of methods for the identification and classification of CVD. Because CVD is a vision impairment, it is crucial to determine if its presence can be predicted with eye behavior. An experiment was conducted using a driving simulator and eye-tracking glasses. The experiment included 27 people with CVD (12 with deuteranopia, 9 with deuteranomaly, 3 with protanopia, and 3 with protanomaly) and 10 people with normal color vision. Each participant performed multiple driving attempts with color-coded guidelines on the navigator to assess visual search ability. Based on data recorded by an eye tracker, the following three types of cross-validated models were developed: binary classification (presence and absence of CVD), a three-class model to recognize the absence of CVD, protanopia, and deuteranopia, and a five-class model to predict the absence of CVD, deuteranopia, deuteranomaly, protanopia, and protanomaly. These three models yielded respective accuracy levels of over 95%, 66%–78%, and 43%–52%. Overall, it was found that eye-tracking metrics have the potential to classify and predict CVD.