An Intelligent Learning Analytics Model for Early Identification of At-Risk Students in Online and Blended Learning
Main Article Content
Abstract
The ability to detect high-risk students who are likely to fail academically or drop out is an urgency in online and blended course set-ups. This research paper is an analytical exploration of early risk identification of learning analytics based on the use of secondary data, and without the creation of a new predictive model or a learning analytics system. Data published and anonymised about Coursera, edX and The Open University are studied to investigate how well academic risk can be detected timely with the help of established AI-assisted learning analytics processes. The analysis is performed in a systematic analytical process, which includes data preprocessing, engagement and assessment indicator construction, analysis of temporal behaviours, at-disengagement pattern detection, and identification of at-risk learning events. Learning analytics literature has suggested analytical procedures that are implemented at an analytical level. Back-testing of potential risk early detection is carried out through historical history back-testing by using known academic performance. The findings suggest that the prevalent factors of academic risk are sustained disengagement and decreasing learning activity. An at-risk learning episode is identified in 312 times out of which 84.6% are identified before the academic outcome is certain. Mean learning periods between early-risk signals and course failure / noncompletion are on average 2.6, indicating that it is possible to obtain meaningful lead time on academic intervention. The level of risk also indicates that high-risk and severe-risk learning conditions are clustered in the middle and later levels of the course development. By and large, the results indicate that AI-intelligent analytical learning analytics processes, used with secondary data, offer a viable and scalable method of the early detection of at-risk students in online and blended educational settings without the need to create new models or roll out new systems.