Classical Machine Learning approaches to detect ADHD from eye movements - Master thesis - Spring 2025

Aug 3, 2026

This master thesis investigates classical machine learning approaches for detecting Attention Deficit Hyperactivity Disorder (ADHD) using eye-movement data. The work covers data collection, preprocessing, visualization, and feature extraction from gaze trajectories and related eye-tracking signals, including variance inflation factor (VIF) analysis for feature selection and handling of multicollinearity. Multiple supervised models are trained and evaluated, including logistic regression, decision trees, random forests, gradient boosting, naive Bayes, and support vector machines, with additional experiments excluding meta-features. Results are compared using model performance and model-selection criteria such as AIC and BIC, followed by discussion of interpretation, implications, limitations, and future directions.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages110

Reading Time4 h 37 min

AccessFree to download

Contents

Acknowledgement

Abstract

1 Introduction

2 Background and State of the Art

3 Data Description and Experimental Methodology

4 Results

5 Discussion and Conclusions

Appendix A Feature Lists and Plots

Bibliography

Classical Machine Learning approaches to detect ADHD from eye movements Syeda Farzana Zerin Thesis submitted for the degree of Master in Applied Computer and Information Technology (ACIT 5900) 30 credits Department of Computer Science Faculty of Technology, Art and Design OSLO METROPOLITAN UNIVERSITY Spring 2025 Classical Machine Learning approaches to detect ADHD from eye movements Syeda Farzana Zerin © 2025 Syeda Farzana Zerin Classical Machine Learning approaches to detect ADHD from eye movements [http://www.oslomet.no/](http://www.oslomet.no/) Printed: Oslo Metropolitan University Contents Acknowledgement ix Abstract xi 1 Introduction 1 2 Background and State ofthe Art 5 2.1 Attention Deficit and Hyperactivity Disorder ................... 5 2.2 Symptoms of ADHD in Children and Adults ................... 6 2.3 How ADHD is Diagnosed ............................ 7 2.4 Eye-tracking technologies and data ........................ 8 2.5 Machine Learning methods: a brief overview .................. 11 2.5.1 Supervised Learning ........................... 11 2.5.2 Unsupervised Learning .......................... 11 2.5.3 Regression/Classifications ........................ 12 2.5.4 Tree-based methods ........................... 13 2.5.5 Support Vector Machine(SVM) ..................... 15 2.6 State of the Art .................................. 16 3 Data Description and Experimental Methodology 21 3.1 Data Collection .................................. 21 3.2 Data Preprocessing and Visualization ...................... 23 3.3 Methodology ................................... 25 3.3.1 Feature Extraction ............................ 26 3.3.2 Variance Inflation Factor (VIF) ...................... 34 4 Results 35 4.1 Visualization of features ............................. 35 4.2 VIF on Features .................................. 36 4.3 Logistic Regression ................................ 37 4.4 Decision Trees .................................. 42 4.5 Random Forests .................................. 45 4.6 Gradient Boosting ................................. 47 4.7 Naive Bayes .................................... 48 4.8 Support Vector Machine ............................. 50 4.9 Comparison of Model Performance ........................ 51 4.10 AIC and BIC in Model Selection ......................... 52 4.11 Model Parameters ................................. 53 4.12 Results Excluding Metafeatures ......................... 54 5 Discussion and Conclusions 61 5.1 Summary of Key Findings ............................ 61 5.2 Interpretation and Implications .......................... 63 5.3 Limitations and Future Directions ........................ 64 Appendix A Feature Lists and Plots 65 Bibliography 93 List of Figures 2.1 An illustration of a basic eye-tracking system [22] ................ 8 2.2 A regression example [33] ............................. 13 2.3 A classification example [34] ........................... 14 3.1 Gaze trajectories (left) and fixation-saccade patterns (right) of an ADHD participant (top) and TD participant (bottom) ................... 24 3.2 Gaze trajectories after outlier removal (left) and blink removal (right) for an ADHD participant (top) and TD participant (bottom) ............... 25 3.3 Flowchart of the ADHD detection methodology.................. 27 3.4 Angle between gaze points............................. 28 3.5 Angle histogram of a sample ADHD (top) and TD (bottom) participant..... 29 3.6 Log velocity histogram of a sample ADHD (top) and TD (bottom) participant.. 30 3.7 Pupil histogram of a sample ADHD (top) and TD (bottom) participant...... 31 3.8 Cross-correlation between log velocity and pupil size of a sample ADHD (top) and TD (bottom) participant............................ 32 3.9 Horizontal or vertical movement dominance using using cos (2θ) ........ 33 4.1 The feature set for A

FAQs

What data source and representation are used to detect ADHD in this thesis?

The thesis uses eye-tracking data, including gaze trajectories and fixation-saccade patterns, and derives multiple features from signals such as angles, velocity, and pupil-related measures.

Which machine learning models are evaluated for ADHD detection?

The study evaluates logistic regression, decision trees, random forests, gradient boosting, naive Bayes, and support vector machines, and compares their performance.

How does the thesis select or refine features before training the models?

Feature selection and refinement are performed using preprocessing steps plus variance inflation factor (VIF) analysis to address multicollinearity, and additional results are generated excluding meta-features.

Aug 3, 2026

Upload by Seraphina