🎓 Student Engagement Detection System

AI-Based Student Engagement Detection in Online Classes Using Optimized MobileNetV2

📌 Project Overview

Student engagement is a critical factor influencing learning outcomes in online education, yet it is difficult to monitor in virtual environments. This project presents an AI-based student engagement detection system that analyzes facial expressions to classify student engagement levels in real time.

The system leverages transfer learning with MobileNetV2, applies multiple optimization techniques, and compares performance against a custom CNN model. Experimental results demonstrate that the optimized MobileNetV2 model achieves the highest accuracy while maintaining computational efficiency suitable for real-time educational platforms.

🎯 Objectives

• Automatically detect student engagement during online classes
• Compare baseline and optimized deep learning models
• Evaluate performance using standard classification metrics
• Address ethical and practical challenges in AI-based education systems

🧠 Engagement Categories

The system classifies students into the following six engagement-related categories:

• Engaged
• Confused
• Frustrated
• Bored
• Drowsy
• Looking Away

🧪 Models Implemented

🔹 Baseline MobileNetV2

• Transfer learning with frozen base layers
• Lightweight and efficient architecture

🔹 Optimized MobileNetV2 (Best Model)

• Enhanced dense layers
• Batch normalization
• Dropout regularization
• Tuned learning rate

🔹 Custom CNN

• Traditional convolutional neural network architecture
• Implemented for comparative performance analysis

📊 Results Summary

Model	Accuracy (%)
Baseline MobileNetV2	94.58
Optimized MobileNetV2	95.68
Custom CNN	93.25

The optimized MobileNetV2 demonstrated improved generalization, reduced overfitting, and superior performance across precision, recall, and F1-score metrics.

⚙️ Technologies Used

• Programming Language: Python
• Frameworks & Libraries: TensorFlow, Keras
• Deep Learning Models: MobileNetV2, CNN
• Data Processing: NumPy, OpenCV
• Visualization: Matplotlib, Seaborn
• Development Platform: Google Colab

📥 Dataset Information

This project uses the Student Engagement Facial Expression Dataset obtained from Kaggle.

⚠️ Due to licensing restrictions, the dataset is not included in this repository.

Dataset download instructions and details can be found in:

dataset/README.md

▶️ How to Run the Project

1️⃣ Clone the Repository

2️⃣ Extract the zip file of Dataset

3️⃣ Place the dataset to the appropriate directory

4️⃣ Run the Notebooks

Open Google Colab and run:

• StudentEngagementDetection_v1.ipynb
• StudentEngagementDetection_v2.ipynb

⚠️ Ethical & Practical Considerations

• Data Privacy: User consent and secure data storage are essential
• Bias & Fairness: Balanced datasets and regular bias audits are required
• Deployment Challenges: Hardware limitations and internet reliability
• Responsible AI: Ethical handling of facial data in educational environments

🎓 Academic Context

• Course: Artificial Intelligence
• Semester: 3rd
• Assignment Type: Project-Based Learning (PBL)
• Department: Software Engineering

🔮 Future Enhancements

• 🎧 Multimodal engagement detection (audio + gaze)
• 🔍 Explainable AI (XAI) for transparency
• 🔐 Federated learning for privacy preservation
• 🏫 Real-time deployment in smart classrooms

👤 Author

• MUHAMMAD HASSAN
• Software Engineering Student
• Batch 2024 – Fall 2025