A Comparative Study of Classification of Human-Written Text Versus AI-Generated Text

The presence of AI in our day-to-day lives has increased very rapidly. There is a need to identify whether a text is human-generated, or AI-generated to ensure its integrity, correctness, and accountability. Many a time, students are using AI-generated texts for their assignments, which questions the integrity and lack of accountability of their work. So, we find the need to classify if a piece of text is generated by AI or is written by a human. We generated the text by training on large data and then generating the next possible character (Fatima et al. in IEEE Access 10:53,490–53,503, 2022). This causes bursts of similar words across generated texts but Human-written text can be very random in nature. We used different classification algorithms to distinguish a text. This proposed model is evaluated using a dataset consisting of 1018 articles, evenly split between AI-generated and human-written texts. Here, we explore the efficiency of various classifiers including KNN, random forest, decision trees, SVC, NuSVC, AdaBoost classifier, gradient boosting classifier, LDA, and QDA. We compare their actuaries to ascertain which one yields the best results for our problem. NuSVC is found to be the most accurate in classifying our test dataset. Our findings contribute to the ongoing research work in this field and provide valuable insights to improve the accuracy of the algorithms employed and choose the most effective out of them.