An Ensemble Learning Approach Based on TabNet and Machine Learning Models for Cheating Detection in Educational...

Area of Science:

• Educational Technology
• Machine Learning
• Data Science

Background:

• Academic dishonesty is a significant challenge in education.
• Machine learning is increasingly used to detect cheating, but deep learning models like TabNet are under-explored.
• Evaluating diverse models is crucial for identifying effective cheating detection methods.

Purpose of the Study:

• To compare the effectiveness of 12 machine learning models, including TabNet, for detecting academic dishonesty.
• To investigate the potential of deep neural networks for tabular data tasks in educational settings.
• To develop and evaluate a hybrid ensemble model for enhanced cheating detection.

Main Methods:

• A comparative analysis of 12 base models: naive Bayes, linear discriminant analysis, Gaussian process, support vector machine, decision tree, random forest, Extreme Gradient Boosting (XGBoost), AdaBoost, logistic regression, k-nearest neighbors, multilayer perceptron, and TabNet.
• Performance evaluation using the area under the receiver operating characteristic curve (AUC).
• Development of an ensemble model by combining the top-performing TabNet and AdaBoost models.

Main Results:

• TabNet demonstrated superior performance (AUC = 0.85) compared to the other 11 base models.
• The hybrid TabNet-AdaBoost ensemble model achieved a significantly higher AUC (0.92).
• Deep neural network models show strong aptitude for tabular data challenges like academic dishonesty detection.

Conclusions:

• TabNet is a highly effective model for detecting cheating in educational tests.
• Ensemble methods, specifically TabNet-AdaBoost, offer improved accuracy for academic dishonesty detection.
• This research provides new insights into applying deep learning for educational integrity.