Time-on-Task Estimation by data-driven Outlier Detection based on Learning Activities

Temporal analysis has been demonstrated to be relevant in Learning Analytics research, and capturing time-on-task, i.e., the amount of time spent by students in quality learning, as a proxy to model learning behaviour, predict performance, and avoid drop-out has been the focus of a number of investigations. Nonetheless, most studies do not provide enough information on how their data were prepared for their findings to be easily replicated, even though data pre-processing decisions have an impact on the analysis’ outcomes and can lead to inaccurate predictions. One of the key aspects in the preparation of learning data for temporal analysis is the detection of anomalous values of temporal duration of students’ activities. Most of the works in the literature address this problem without taking into account the fact that different activities can have very different typical execution times. In this paper, we propose a methodology for estimating time-on-task that starts with a well-defined data consolidation and then applies an outlier detection strategy to the data based on a distinct study of each learning activity and its peculiarities. Our real-world data experiments show that the proposed methodology outperforms the current state of the art, providing more accurate time estimations for students’ learning tasks.