Detecting generative artificial intelligence essays using large language models: Machine and deep learning approaches

Abstract

The study focuses on discerning between human and AI-generated essays, highlighting the ethical implications of AI in academia. It employs various algorithms like logistic regression, Support Vector Machine (SVM), decision trees, random forests, KNN, and LSTM to develop models for essay classification. The TF-IDF technique (Term Frequency-Inverse Document Frequency) is applied to assess document word importance, with rigorous parameter tuning ensuring model accuracy. Findings revealed SVM's exceptional precision and recall, highlighting its robustness in accurately classifying essays, while decision trees offer simplicity but increased misclassification risk. KNN strikes a balance and random forests as well. LSTM excels in contextual understanding, albeit with higher computational demands. The research emphasizes the significance of algorithm selection in maintaining academic integrity and fostering genuine student creativity. SVM emerges as a robust and accurate choice for essay classification, ensuring fair assessment and upholding academic honesty.