Detection of cognitive and attention dimensions in block programming interface for learning sensor data analytics in construction education
•Learners’ focus and difficulty in block-based coding are detected via eye-tracking.•Ensemble model outperforms other models on seven eye movement features.•Gaze position and pupil diameter emerge as key predictors.•Implications, challenges, and strategies for bridging skill gaps are discussed. The...
Saved in:
| Published in: | International journal of human-computer studies Vol. 205; p. 103626 |
|---|---|
| Main Authors: | , , , |
| Format: | Journal Article |
| Language: | English |
| Published: |
Elsevier Ltd
01.11.2025
|
| Subjects: | |
| ISSN: | 1071-5819 |
| Online Access: | Get full text |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | •Learners’ focus and difficulty in block-based coding are detected via eye-tracking.•Ensemble model outperforms other models on seven eye movement features.•Gaze position and pupil diameter emerge as key predictors.•Implications, challenges, and strategies for bridging skill gaps are discussed.
The increasing adoption of sensing technologies in the construction industry generates vast amounts of raw data, requiring analytics skills for effective extraction, analysis, and communication of actionable insights. To address this, ActionSens, a block-based programming interface, was developed to equip undergraduate construction engineering students with domain-specific sensor data analytics skills. However, efficient user interaction with such tools requires integrating intelligent systems capable of detecting users’ attention and cognitive states to provide context-specific and tailored support. This study leveraged eye-tracking data from construction students during the usability evaluation of ActionSens to explore machine learning models for classifying areas of interest and interaction difficulties. For visual detection, key interface elements were defined as areas of interest, serving as ground truth, while interaction difficulty was labeled based on participant feedback for reported challenges. The Ensemble model demonstrated the highest performance, achieving 88.3% accuracy in classifying areas of interest with raw data, and 82.9% for classifying interaction difficulties using oversampling techniques. Results show that gaze position and pupil diameter were the most reliable predictors for classifying areas of interest and detecting interaction difficulties. This study pioneers the integration of machine learning and eye-tracking with block-based programming interfaces in construction education. It also reinforces the Aptitude-Treatment Interaction theory by demonstrating how personalized support can be adapted based on individual cognitive aptitudes to enhance learning outcomes. These findings further contribute to the development of adaptive learning environments that can detect specific user aptitudes and provide context-specific guidance, enabling students to acquire technical skills more effectively. |
|---|---|
| ISSN: | 1071-5819 |
| DOI: | 10.1016/j.ijhcs.2025.103626 |