Related Experiment Video
Updated: Jan 16, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.3K
Exploratory machine learning models for computer vision syndrome in occupational health
Magda Isabel Sebinha1, Ana Paula Oliveira1,2, Clara Martinez-Perez1
1Instituto Superior de Educação e Ciências de Lisboa (ISEC Lisboa), Lisboa, Portugal.
Digital Health
|September 29, 2025
Summary
Predictive models show promise for identifying computer vision syndrome (CVS) risk in workers. Machine learning, particularly XGBoost, and logistic regression achieved the highest accuracy, highlighting the importance of demographic and behavioral factors.
Area of Science:
- Occupational Health
- Digital Health
- Machine Learning Applications
Background:
- Computer Vision Syndrome (CVS) is an increasing occupational health issue due to extensive digital device use.
- A scarcity of robust predictive models hinders effective risk identification for CVS in the workplace.
- The CVS-Q© questionnaire serves as the gold standard for defining and assessing CVS cases.
Purpose of the Study:
- To evaluate and compare traditional and machine learning (ML) predictive models for Computer Vision Syndrome (CVS).
- To assess the accuracy of these models in identifying CVS cases based on the CVS-Q© questionnaire.
- To identify key predictors of CVS within an occupational setting.
Main Methods:
- A cross-sectional study involving 90 Portuguese workers regularly using digital display devices.
- Data collection via self-administered questionnaires on demographics, refractive errors, screen time, and CVS symptoms (CVS-Q©).
- Development and comparison of logistic regression and ML models (RF, GBM, XGBoost, SVM) with SMOTE augmentation and 5-fold cross-validation.
Main Results:
- Logistic regression and XGBoost achieved the highest accuracy (70.6%) and discriminative ability (AUC 0.773 and 0.758).
- All models showed high sensitivity (≥72.7%), but specificity varied (16.7%–50%).
- Female sex was a significant predictor in logistic regression; sex, screen time, and age were key in ML models; refractive errors had minimal impact.
Conclusions:
- Predictive modeling, especially using demographic and behavioral variables, shows potential for occupational CVS risk identification.
- The study's preliminary findings require validation in larger, more representative cohorts.
- Further research is necessary to confirm the scalability and clinical utility of these predictive models.
