Differences in learners' behavior have a deep impact on their educational performance. Consequently, there is a need to detect and identify these differences and build suitable learner models accordingly. In this paper, we report on the results from an alternative approach for dynamic student behavioral modeling based on the analysis of time-based student-generated trace data. The goal was to unobtrusively classify students according to their time-spent behavior. We applied 5 different supervised learning classification algorithms on these data, using as target values (class labels) the students' performance score classes during a Computer-Based Assessment (CBA) process, and compared the obtained results. The proposed approach has been explored in a study with 259 undergraduate university participant students. The analysis of the findings revealed that a) the low misclassification rates are indicative of the accuracy of the applied method and b) the ensemble learning (treeBagger) method provides better classification results compared to the others. These preliminary results are encouraging, indicating that a time-spent driven description of the students' behavior could have an added value towards dynamically reshaping the respective models.
LA/EDM research results indicate that data integration from multiple sources can improve the accuracy of a learner profile and subsequent adaptation and personalization of content. Exploration of students’ behavior within educational contexts that support multimodality and mobility could lead to shaping a holistic picture of how, when and where learning occurs.
The paper reports an examination of the literature on experimental case studies conducted in the domain from 2008 to 2013. Although the search terms identified 209 mature pieces of research work, the inclusion criteria limited the key studies to 40. The research questions, methodology and findings of these published papers were analysed and categorized. The authors used non-statistical methods to evaluate and interpret findings of the collected studies. Their results highlighted four distinct major directions for learning analytics / educational data mining empirical research. The paper also discusses the value added by this research
The social dimension of learning and the opportunity of selectively participating in MOOCs are also explored in research with encouraging results. Consequently, the research community could gain insight into the learning mechanisms that previously were a “black box.”
Researchers set the educational context within limits in which previously it was almost impossible to infer behavior patterns, due to their high levels of granularity. In such advanced learning contexts, LA/EDM research community determines simple and/or sophisticated factors as predictors of performance and explores their predictive value and capabilities by tracking actual data and changes on behavioral data. The goal is to identify the most significant factors in order to develop better systems. These systems will allow students to monitor their own progress and will help them evaluate and adjust their learning strategies to improve their performance in terms of learning outcomes.
Abstract: Predicting student’s performance is a challenging, yet complicated task for institutions, instructors and learners. Accurate predictions of performance could lead to improved learning outcomes and increased goal achievement. For that reason, prediction of performance is acknowledged as one of the major objectives of Learning Analytics and Educational Data Mining research. In this paper we explore the predictive capabilities of student’s time-spent on answering (in-)correctly each question of a multiple-choice assessment quiz, along with student’s final quiz-score, in the context of computer-based testing. We also explore the correlation between the time-spent factor (as defined here) and goal-expectancy. We present a case study and investigate the value of using this parameter as a learning analytics factor for improving prediction of performance during computer-based testing. Our initial results are encouraging and indicate that the temporal dimension of learning analytics should be further explored.
Overview from the Hub: The preliminary results from this research indicate that temporal learning analytics – that is to say the cumulative duration of time that a student takes to make correct or incorrect answers on a multiple choice quiz - have a statistically significant capability of predicting actual performance (almost 62% of the variance). In their single case study of 96 high school students the report finds that if a learner spends more time to answer correctly then they are more likely to score higher (positive effect of 0.60) and if they spend more time answering incorrectly then it appears they are likely to score lower (0.29). They also report a third finding: that there is a positive effect of (learner pre-stated) goal expectancy on correct answers and a negative effect on the total time spent making incorrect answers. The proposed use of this technique once researched further is as a potential replacement for traditional grading activities where determining grades act as effort overload.