Abstract
Learning confusion is a common emotion among learners. With the aid of machine learning, this paper develops a data-driven emotion model that automatically recognizes learning confusion in facial expression images. The data on learning behaviors and learning confusion of multiple subjects were collected through an online English evaluation experiment, and imported to the proposed model to derive the relationship between learning confusion and academic performance, which is measured by the correctness of the students’ answers to the test questions. The experimental results show that the students with learning confusion had relatively low correct rate of answering test questions. The research findings reveal the relationship between learning confusion and academic performance, laying the basis for predicting the academic performance of English learners through machine learning.
Introduction
In the landscape of online education, instructors often lack the immediate visual feedback necessary to gauge student comprehension. "Learning confusion" is a pivotal emotional state; while transient confusion can facilitate cognitive breakthroughs, persistent confusion often leads to frustration, disengagement, and academic decline. This research bridges the gap between computer vision and educational psychology by identifying confusion through facial recognition and quantifying its direct impact on a student's ability to solve problems correctly. The ultimate goal is to enable intelligent teaching systems that can provide real-time interventions when a student appears lost.
Methods and Technical Roadmap
The researchers implemented a multi-stage technical roadmap to transition from raw visual data to actionable academic predictions.
Data Acquisition and Preprocessing
The study involved 200 students (100 male, 100 female) participating in an online English test. A camera captured 30,000 facial expression images at 640×480 resolution. During preprocessing, "noisy" data—such as instances where students moved out of the frame—were discarded, leaving 25,080 valid datasets. From these, 78 key facial feature points (including corners of the mouth, eyes, and eyebrows) were extracted to create a 156-dimensional dataset.
Model Construction
Six machine learning architectures were evaluated to determine the most effective method for distinguishing "confusion" from "no confusion":
- Logistic Regression (LR)
- Support Vector Machine (SVM)
- k-Nearest Neighbors (KNN)
- Decision Tree (DT)
- Random Forest (RF)
- Deep Feedforward Neural Network (DFNN)
As illustrated in the Random Forest architecture, the model employs bootstrap sampling to create multiple training sets, which are then processed by individual decision trees. The final classification result is determined through a voting mechanism, enhancing the model's robustness and accuracy.
Key Results and Insights
Performance of Classification Models
Surprisingly, the Random Forest (RF) classifier emerged as the top performer with an accuracy of 71.39%, surpassing the Deep Feedforward Neural Network (65.89%). This indicates that for feature-engineered facial datasets of this scale, ensemble-based traditional machine learning can be more effective and computationally efficient than deep learning. Logistic Regression was the least effective, achieving only 58.79% accuracy.
The Confusion-Performance Link
The study established a clear negative correlation between confusion and academic success. Students flagged as "confused" demonstrated significantly lower correct answer rates. When using Logistic Regression to predict test correctness based solely on the presence of confusion, the model reached 62.95% accuracy.
Behavioral Observations
Behavioral data revealed that confusion often triggers proactive help-seeking. In a sample of 1,000 questions, 245 were marked as confusing; in 183 of those instances, students actively clicked the "analysis" button to seek an explanation. Interestingly, predicting academic performance using only the "learning confusion" variable was more accurate (62.95%) than combining it with behavioral data like "checking analysis" (56.71%), suggesting that emotional state is a purer indicator of understanding than subsequent corrective actions.
Limitations and Future Work
Several challenges remain for real-world application. Physical noise, such as head movements, rendered nearly 17% of the captured images unusable, necessitating more robust tracking systems. Additionally, while the model successfully identifies confusion in English learners, further research is required to determine if these facial markers are universal across different subjects (e.g., mathematics) or cultural backgrounds. The logical next step is the development of a real-time system that can intervene—by simplifying content or offering hints—the moment confusion is detected.
Summary updated