Yaosheng Lou

Predicting student performance has been a critical focus of educational research. With an effective predictive model, schools can identify potentially at-risk students and implement timely interventions to support student success. Recent developments in educational data mining (EDM) have introduced several machine learning techniques that can effectively analyze students’ demographic information, learning processes, and other contextual factors to predict academic outcomes. However, limited research has compared the predictive accuracy of these EDM techniques with traditional statistical methods in real-world educational settings. This case study aims to address this gap by empirically evaluating the performance of generalized linear regression, decision tree, and random forest regression in predicting three end-of-course exams. The data are from a statewide high school dataset. Model performance was assessed using R-square, RMSE, MAE, and MSE. The results indicated that generalized linear regression consistently outperformed decision tree and random forest regression in terms of both predictive accuracy and error. Additionally, this study examined the capacity of these methods to identify important predictors. These findings may offer valuable insights for researchers and educators in selecting appropriate methods for similar prediction tasks.

Despite the numerous benefits of technology, the risk of over-reliance on technology has been noticed and emphasized by many scholars in recent decades. If users depend on technology heavily, the excepted benefits, for example, reducing errors, increasing efficiency, and improving safety, might not always be realized and can be offset. However, existing research in over-reliance on the academic database, one of the common used technology products, is particularly limited. This case study investigated how a well-known database, Web of Science, misled researchers' work. While academic databases provide many literature searching opportunities, academic researchers should realize that databases may also have technical errors. The authors introduced two possible explanations (i.e., "automation bias and complacency" and "the theory of technology dominance") to explain how this phenomenon occurred in the scholarly community. In the end, the article is devoted to offering some feasible recommendations for both users and database developers.