Table 1.
Summary of studies using eye features.
| References | Year | Topic domain | Objective | Eye features | Subjects | Eye-trackers | Classifier | Performance |
|---|---|---|---|---|---|---|---|---|
| Cao et al. (2016) | 2016 | Intention recognition | To examine and evaluate whether pupil variation has a relevant impact on the endoscopic manipulator activation judgment | Pupil size, velocity of eye rotation | 12 (10 males, 2 females) | Tobii 1750 | SVM and PNN | 88.6% |
| Ahmed and Noble (2016) | 2016 | Image classification | Attempt to classify and acquiring the image frames of the head, abdominal, and femoral from 2-D B-Mode ultrasound scanning | Fixations | 10 | EyeTribe (30Hz) | Bag of words model | 85–89% |
| Zhang and Juhola (2016) | 2017 | Biometric identification | To study primarily biometric recognition as a multi-class classification process and biometric authentication as binary classification | Saccades | 109 | EyeLink (SR Research) | SVM, LDA, RBF, MLP | 80–90% |
| Zhou et al. (2017) | 2017 | Image classification | To propose an approach of two-stage feature selection for image classification by considering human factors and leveraging the importance of the eye-tracking data. | Fixations, ROI | - | Tobii X120 | SVM | 94.21% |
| Borys et al. (2017) | 2017 | User performance classification in RFFT | To verify and evaluate whether eye-tracking data in combination with machine learning could be used to identify user output in RFFT. | Fixations, saccades, blinks, pupil size | 61 | Tobii Pro TX300 | Quadratic discriminant analysis | 78.7% |
| Karessli et al. (2017) | 2017 | Image classification | To propose an approach that uses gaze data for zero-shot image classification | Gaze point | 5 | Tobii TX300 (300Hz) | SVM | 78.2% |
| Labibah et al. (2018) | 2018 | Lie detection | To construct the object using a lie detector with the analysis of pupil changes and eye movements using image processing and decision tree algorithm. | Pupil diameter, eye movements | 40 | Computer camera | Decision tree | 95% |
| Qi et al. (2018) | 2018 | Material classification | To investigate how humans interpret material images and find information on eye fixation enhances the efficiency of material recognition. | Fixation points, gaze paths | 8 | Eye-tracker | CNN | 85.9% |
| Singh et al. (2018) | 2018 | Reading pattern classification | To analyze the reading patterns of eye-tracking inspectors and assesses their ability to detect specific types of faults. | Fixations, saccades | 39 | EyeLink 1000 | NB, MNB, RF, SGD, ensemble, decision trees, Lazy network | 79.3–94% |
| Lagodzinski et al. (2018) | 2018 | Cognitive activity recognition | To discuss the concept of the eye movement study, which can be used effectively in behavior detection due to the good connection with cognitive activities. | EOG, accelerometer data | 100 | JINS MEME EOG-based eye-tracker | SVM | 99.3% |
| Bozkir et al. (2019) | 2019 | Cognitive load classification | To propose a scheme for the detection of cognitive driver loads in safety-critical circumstances using eye data in VR. | Pupil diameter | 16 | Pupil Labs | SVM, KNN, RF, decision trees | 80% |
| Orlosky et al. (2019) | 2019 | User understanding recognition | To recognize the understanding of the vocabulary of a user in AR/VR learning interfaces using eye-tracking. | Pupil size | 16 | Pupil Labs Dev IR camera | SVM | 62–75% |
| Sargezeh et al. (2019) | 2019 | Gender classification | To examine parameters of eye movement to explore gender eye patterns difference while viewing the indoor image and classify them into two subgroups. | Saccade amplitude, number of saccades, fixation duration, spatial density, scan path, RFDSD | 45 (25 males, 20 females) | EyeLink 1000 plus | SVM | 84.4% |
| Tamuly et al. (2019) | 2019 | Image classification | To develop a system for classifying images into three categories from extracted eye features. | Fixation count, fixation duration average, fixation frequency, saccade count, saccade frequency, saccade duration total, saccade velocity total | 25 | SMI eye-tracker | KNN, NB, decision trees | 57.6% |
| Luo et al. (2019) | 2019 | Object detection | To develop a framework for extracting high-level eye features from low-cost remote eye-tracker's outputs with which the object can be detected. | Fixation length, radius of fixation, number of time-adjacent clusters | 15 (6 males, 9 females) | Tobii Eye Tracker 4C | SVM | 97.85% |
| Startsev and Dorr (2019) | 2019 | ASD classification | To propose a framework that identifies an individual's viewing activity as likely to be correlated with either ASD or normal development in a fully automated fashion, based on scan path and analytically expected salience. | Fixations, scan path | 14 | Tobii T120 | RF | 76.9% AUC |
| Zhu et al. (2019) | 2019 | Depression recognition | To propose a depression detection using CBEM and compare the accuracy with the traditional classifier. | Fixation, saccade, pupil size, dwell time | 36 | EyeLink 1000 | CBEM | 82.5% |
| Vidyapu et al. (2019) | 2019 | Attention prediction | To present an approach for user attention prediction on webpage images. | Fixations | 42 (21 males, 21 females) | Computer webcam | SVM | 67.49% |
| Kacur et al. (2019) | 2019 | Schizophrenia disorder detection | To present a method to detect schizophrenia disorder using the Rorschach Inkblot Test and eye-tracking. | Gaze position | 44 | Tobii X2-60 | KNN | 62% - 75% |
| Yoo et al. (2019) | 2019 | Gaze-writing classification | To propose a gaze-writing entry method to identify numeric gaze-writing as a hands-free environment. | Gaze position | 10 | Tobii Pro X2-30 | CNN | 99.21% |
| Roy et al. (2017) | 2020 | Image identification | To develop a cognitive model for ambiguous image identification. | Eye fixations, fixation duration, pupil diameter, polar moments, moments of inertia | 24 (all males) | Tobii Pro X2-30 | LDA, QDA, SVM, KNN, decision trees, bagged tree | ~90% |
| Guo et al. (2021) | 2021 | Workload estimation | To investigate the usage of eye-tracking technology for workload estimation and performance evaluation in space teleoperation | Eye fixation, eye saccade, blink, gaze, and pupillary response | 10 (8 males, 2 females) | Pupil Labs Core | LOSO protocol, SVM (RBF) | 49.32% |
| Saab et al. (2021) | 2021 | Image classification | To propose an observational supervision approach for medical image classification using gaze features and deep learning | Gaze data | - | Tobii Pro Nano | CNN | 84.5% |