At-risk student identification and interventions for data science programs at a South African university

In this paper, thresholds are established to identify at-risk data science students at a South African university and an intervention process is proposed for handling identified at-risk students. An evaluation of student performance in the core program modules is conducted, focusing on the differences between the standard and extended data science programs offered by this university. Through this evaluation, mark thresholds are specified for core mathematics and statistics modules that can be used to detect at-risk students. A statistical analysis is conducted to determine the suitability of using the thresholds for identifying at-risk students. A fitted logistic regression model, using the number of threshold breaches as the predictor, yields significant predictor coefficients and odds ratios for both programs ( p = 0.0014 and OR = 4.0367 for the standard program; p = 0.0405 and OR = 2.1174 for the extended program). For both programs, the Mann–Whitney test confirms a statistically significant difference in the number of threshold breaches between graduates and dropouts ( p < 0.0001; p = 0.0273) and Fisher’s exact test indicates an association between the number of breaches and dropout status ( p = 0.0002; p = 0.0312). Lastly, sensitivity/specificity analysis using the number of breaches to classify students yields estimated AUC values of 0.7811 and 0.7074, respectively. An intervention process is also suggested for the data science programs to provide struggling students with advice throughout their academic life cycles. This study shows how a simple threshold approach can be used to design an understandable and program-specific at-risk identification strategy. Literature on extended programs is less common than literature on bridging programs, where the differences between these transition programs are also highlighted in this paper.