Analysing the predictive power for anticipating assignment grades in a massive open online course

Abstract

The learning process in a MOOC (Massive Open Online Course) can be improved from knowing in advance learners' grades on different assignments.This would be very useful to detect problems with enough time to take corrective measures. In this work, the aim is to analyse how different course scores can be predicted, what elements or variables affect the predictions, and how much and in which way it is possible to anticipate scores. To do that, data from a MOOC about Java programming have been used. Results show the importance of indicators over the algorithms and that forum-related variables do not add power to predict grades, unlike previous scores. Furthermore, the type of task can vary the results. Regarding the anticipation, it was possible to use data from previous topics but with worse performance, although values were better than those obtained in the first seven days of the current topic.

Generalizing Predictive Models of Admission Test Success Based on Online Interactions

“…However, in this case, results show that the trend is similar with respect to the most important variables and that variables related to exercises are the best predictors. This was also shown in other contexts, such as in the article by Moreno-Marcos et al [4], which entails that research results can be applicable in other contexts, although they may change if they are very different. Because of that, it is important to analyze different contexts and to compare the results to obtain global conclusions about how students learn and about what behaviors have relevant effects on their success, although current results can actually provide insight in effective learning behaviors.…”

“…org/packages/dplyr/versions/0.7.8). These high-level variables are similar to those used in previous contributions (e.g., Reference [4]), and their intent was to gather information about the main kind of features (according to Reference [16]): accesses to the platform, videos, and exercises. In this studio, forum variables are not considered because of the very low level of forum interactions.…”

“…Moreover, these digital platforms not only serve as a repository to upload teaching materials but also allow us to get comprehensive traces about learners' interactions. These interactions can be useful to detect patterns about students' behaviors and to predict trends on advance (e.g., who will pass the course) [4].…”

Analysis of the Factors Influencing Learners’ Performance Prediction With Learning Analytics

“…Among variables related to forum, some examples, used by Klüsener and Fortenbacher [35], are the number of messages, number of words in the messages, and number of ratings emitted/received. However, they are opposite findings about their effectiveness in the predictive models of performance as forum variables can be useful in some contexts (e.g., [36]) but not in others (e.g., [5]). For example, [36] found that variables related to the quality/content of the posts, such as the length of posts, were useful predictors for dropout, while forum variables were not good predictors at all in the research by Moreno-Marcos et al [5] when predicting assignment grades (even using the same variables).…”

“…Among those factors, the variables used to develop the models and the way they are collected (e.g., if variables are collected from the beginning of the course in a cumulative way or if they are collected in a specific period) can be very relevant. As suggested by Moreno-Marcos et al [5], selection of data can be even sometimes more important than the algorithms because variables need to capture appropriate information in relation to the variable to be predicted. Because of that, it is worth analyzing the predictive power of different sets of features.…”