Understanding Student Learning Behavior and Predicting Their Performance

Understanding Student Learning Behavior and Predicting Their Performance

Muhammad Wasif (Information Technology University, Pakistan), Hajra Waheed (Information Technology University, Pakistan), Naif R. Aljohani (Faculty of Computing and Information Technology, King Abdulaziz University, Saudi Arabia) and Saeed-Ul Hassan (Information Technology University, Pakistan)
Copyright: © 2019 |Pages: 28
DOI: 10.4018/978-1-5225-9031-6.ch001
OnDemand PDF Download:
No Current Special Offers


Despite the increase in the adoption of online educational platforms, student retention is still a challenging task with a number of students having low performance margins during these courses. This chapter intends to predict student performance based on their learning behavior on the basis of their logging data history, using the publicly available Open University Learning Analytics Dataset. To model this problem, logistic regression (LR) is used as a baseline technique. Additionally, random forest (RF), multiple layered perceptron with multiple activation functions, and Gaussian Naïve Bayes are also deployed. The results demonstrate that RF outperforms the baseline LR and other models with 89% accuracy, 89% precision, 88% recall, and 88% F1-score. Finally, the authors conclude that using the above-mentioned models, students “at-risk” can be identified which can be managed by an alert mechanism to improve student success rate by making timely interventions.
Chapter Preview


The richness of online educational data has opened a new area of interest for researchers to understand the learning behavior and predicting the performance of learners by deploying various machine learning techniques. In the research community, many studies have demonstrated interest in predicting learners’ interaction pattern and their corresponding performances from the available educational data of online learning platforms. Researchers are trying to make statistically based early interventions using available data to predict students at-risk of failing the course and those that are more likely to yield better performances. In the past few years, online teaching and learning has gained more popularity all around the world especially in countries like U.S and Australia, where Learning Management Systems (LMS) have gained more attraction. These tools generate large amount of students’ interaction data which can be used to predict their likely outcomes. At first hand Data Mining (DM) was used to analyze prospects of educational data using prediction techniques such as classification and association.

Educational Data Mining (EDM) has become an interesting area of research to extract the hidden information and patterns from the data. These inspirations lead to the establishment of conference series, International Conference on EDM, followed by the establishment of the journal of EDM. Another group of researchers founded the Society for Learning Analytics Research (SoLAR) that established another conference series, International Conference of Learning Analytics and Knowledge (LAK), specifically to understand how student’s data may be collected, analysed and reported to improve students’ learning. Recently, a new journal of LA has been established as a result of these growing research efforts examining the potential of analysing educational data (Waheed et al. 2018).

Educational data generated as a result of users’ interactions with the online learning system is useful in multiple aspects which have led to a multidisciplinary research field. Researchers from different backgrounds across the globe are collaborating and new dimensions such as learners’ analytics, academic analytics, educational data mining and learning analytics (LA), have emerged, related with the analysis of educational data. The commonality in all of these terms is the use of different or similar aspects of educational data for improvising the learning environment. Recently, a new term has been introduced: ‘educational data science’, which clarifies how disciplines and researchers with different research interests and backgrounds can work in this field (Romero & Ventura 2017).

The LA research community lays emphasis on the collection and examination of students’ behavior, deploying techniques to enhance understanding that yield to an optimal environment by improving learner’s performance (Siemens & Long, 2011).These collective measures assist an institute in maintaining a positive educational atmosphere, supporting an institute in maintaining its conduct (Daniel 2017; Romero & Ventura, 2010). Furthermore, along with improving the students learning and the teachers teaching mechanism, LA also assists educational institutions in developing new policies considering students behaviors and strategies using predicted insights. These insights help institutes to take more effective information based decisions which ultimately supports the educational environment to yield positive outcomes (Leitner, Khalil, & Ebner, 2017). Khalil and Ebner (2015) described it as an analysis technique applied to the educational data stream to infer patterns for improving and elevating students’ performance and assisting in teaching mechanisms.

Complete Chapter List

Search this Book: