Search

Learning analytics is a data-centric field that applies machine learning algorithms in the educational domain to analyze e-Learning environment data. This study employs descriptive and predictive learning analytics approaches in order to develop descriptive and predictive models of student behavior and success. Cluster analysis, unsupervised machine learning algorithm, and decision tree, supervised machine learning algorithm, are applied on the data from a Learning Management System (LMS) Moodle. Research results indicated: (i) groups of students with similar patterns in behavior at LMS, (ii) student activities at LMS that lead to successful course completion. Such results serve as guidelines for teachers when developing courses and students when enrolling in the course. Descriptive and predictive learning analytics is an innovative approach in education that can enhance teachers and students and improve learning outcomes.

Education domain generates huge amounts of data, especially in online teaching and learning. Data analysis enables detecting patterns of students' learning behavior which leads to personalized attention and adaptive feedback. Learning management system (LMS) data are in the focus of this paper. The LMS logs store information about students’ login frequency, time of visits, number of downloading different resources, time and frequency of various activities. Within this research, log file analysis is performed from a Business decision making course at the University of Zagreb. Raw LMS data were extracted from Moodle, data was prepared, explained and explored in order to detect patterns in student’s behavior in the online environment. Research results provide a basis for teachers’ interventions on one hand, and serve as an input into predictive models’ development, on the other hand.

Educational institutions tend to use learning management systems (LMS) to enhance teaching and learning process. An LMS collects data about students’ activities in log files. This data is used in prediction of student success. Machine learning algorithms are increasingly used to support development of accurate and reliable descriptive and predictive models from educational data. However, selection of machine learning algorithm is complex task and depends on characteristics of dataset used in the analysis. Characteristics of dataset (meta-features) can explain why one machine learning algorithm performs better on certain dataset than other algoriths. Meta-learning field deals with this issue. Identification of meta-feature in educational domain is still insufficiently researched area. In this paper, we are trying to fill that gap by providing meta-features identification in educational domain. Our focus are general meta-features and learning management system data. Research results indicated patterns in LMS datasets meta-features.

Huge volumes of data created by students` activities on learning management systems (LMS) urged an opportunity to extract meaningful information from data. Development of data mining field yielded algorithms making possible to analyse data with the aim to improve quality of the educational processes. In this paper, we are comparing four data mining methods based on different machine learning algorithms to predict academic performance of IT students based on data about their activities at the LMS. Aim of the research were twofold: (i) to predict students' academic achievement and identify most important predictors of academic success, (ii) to compare different machine learning algorithms and identify which fits the best on LMS data. Research results indicated frequency of students` discussions as the most important predictor of success. Neural networks provided most accurate and reliable model.