Abstract:

Student dropout is a significant issue for higher education institutions, and Educational Data Mining (EDM) enables the extraction of knowledge from educational data. In a critical area like education, it is essential to understand how the result is generated. In this way, eXplainable Artificial Intelligence (XAI) offers solutions that enhance user understanding of AI systems, promoting ethical, transparent, and responsible use. This research aims to analyze the impact of the COVID-19 pandemic on student dropouts and retention at a public higher education institution using a Random Forest model. Additionally, we seek to improve the explainability of the output through an XAI post-hoc method, specifically SHAP. This approach enables data mining in big data, allows us to compare trends before and after the pandemic, and identify significant factors for dropout prediction. Our findings indicate that there has been a change in dropout and retention trends post-pandemic, and the locations of students, campuses, and their families play a significant role in dropout predictions. These factors are dynamic and are influenced by external and internal factors from the higher education journey, such as the pandemic, socioeconomic country characteristics, family, psychological factors, and student generation. Therefore, it is essential to analyze the dataset frequently and split it into periods. Analyzing big data within a university system is crucial for improving student retention, and the present study points out factors in the dataset that need to be further analyzed for understanding the reasons why students leave higher education without completing their courses.

Keywords:

Education data mining, explainable artificial intelligence, artificial intelligence, dropout, retention, higher school education.