More Data is not Always Better Data: An Exploratory Learning Analytics Study in Early Prediction

Abstract

This study aims to explore the early prediction and forecasting of students’ performance using learning analytics methods. We do so by examining a large number of students at four different time points and employing four machine learning algorithms. We seek to fill a gap in the literature regarding the early prediction of students’ performance versus prediction at later time points. The results revealed that the data for the whole course did not consistently outperform earlier time points across all performance indicators. Surprisingly, forecasts based on data from the first week had reasonable accuracy, implying that preemptive interventions can be implemented as early as the first week. Furthermore, predictions made in the second week performed the best, probably due to students’ initial motivation. Decision trees were the best-performing algorithm, consistently displaying acceptable performance across all time points. This study indicates that the implementation of early predictive learning analytics is not only viable but also reliable with reasonable accuracy and that accumulating more data has no added benefits in our case. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024.