The Relationship Between Artificial Intelligence Skills and Academic Performance Before Exams: A Comparative Analysis of SPSS and Keras-Based Artificial Neural Networks

Abstract

This paper aims to examine the various aspects of the use of generative AI tools by undergraduate students during the pre-exam period, including their skill levels with these tools and the impact that this has upon their academic performance - all determined through both traditional statistical methods (via SPSS) and machine learning methods (via a Keras-based MLP). The study is based upon the responses of 1,231 undergraduate students. Statistical analyses performed in SPSS as well as with machine learning models in Python (Keras library) and statistical software (PCA and K-Means models) were performed to arrive at the conclusions of the study.

Results indicate that 94.7% of students use generative AI tools. The most common platforms used by students are ChatGPT (59.8%) and Gemini (33.0%). Furthermore, there was a significant positive correlation (r = 0.64) between a student’s skill levels with AI tools and their self-efficacy in their academic subjects. Finally, the MLP was able to achieve an accuracy rate of 87.4% in the classification of students according to their academic performance.

This survey was performed in 2026 and included 1231 students from Mingachevir State University. The survey was performed only among undergraduate students. Finally, the survey was performed during the May-June 2026 time period, just prior to exams.

In discussing the limitations of commercial AI detection software, the authors of the study found support in their results for the potential limitations of such software. Thus, the authors conclude their paper with a discussion of the importance of AI literacy in relation to academic performance, as well as in the potential to combine the two methods of analysis in future studies and investigations.