PREDICTIVE ANALYTICS IN HIGHER EDUCATION: A COMPARATIVE STUDY OF ARTIFICIAL INTELLIGENCE APPROACHES ACROSS MULTIPLE COHORTS
PREDICTIVE ANALYTICS IN HIGHER EDUCATION: A COMPARATIVE STUDY OF ARTIFICIAL INTELLIGENCE APPROACHES ACROSS MULTIPLE COHORTS

Summary/Abstract: Predicting student success in higher education is critical for enhancing retention, enabling timely interventions, and improving the overall quality of the study process. This paper is about evaluation and comparing several artificial intelligence approaches on the same dataset of cohorts, covering multiple generations and great number of students, including both graduates and non-graduates. A range of AI tools is applied with respect of the known large language models (LLMs) using natural language prompts in order to predict academic outcomes, final grade point average, dropout risk etc. Prompts were designed to identify at-risk students, to show key predictors of success based on data available, as well as compare quality cross generations. Outputs were evaluated in terms of predictive accuracy, interpretability and applicability for decision making processes in higher education. Obtaining reliable information about adding more data into the dataset for more accurate forecasting was also considered. The paper shows that prompt-based AI can generate actionable insights from large datasets from higher education with no extensive coding or specialized technical expertise. Some AI platforms emphasized quantitative precision, whereas others highlighted qualitative explanations and contextual risk factors, illustrating the complementary strengths of different approaches. Cohort-level analysis revealed meaningful trends across all the generations of students that were analyzed, including fluctuation in graduation rates and performance clusters linked to curriculum changes. For sure, the forecasting accuracy is different across different tools, but consistent patterns emerged regarding critical early courses, as well as the important engagement indicators. This study contributes to the growing body of research on AI in higher education by demonstrating a systematic comparison of multiple prompt-based tools on the same dataset. It highlights both the opportunities and limitations of natural-language AI for predictive analytics in higher education, emphasizing the balance between performance, interpretability, and ease of use. The results suggest that combining predictive accuracy with transparent explanations can support more informed interventions, enhance program evaluation, and ultimately improve student outcomes. Furthermore, the study anticipates that different AI tools will not only vary in predictive strength but also offer distinct recommendations for improving forecasting accuracy. These insights are expected to inform future methodological choices and guide the design of more effective, adaptive educational analytics frameworks.

• Details
• Contents