Data and Process Mining in Analysing Student Behaviour

Interdisciplinary Description of Complex Systems 23(5), 467-483, 2025 DATA AND PROCESS MINING IN ANALYSING STUDENT BEHAVIOUR Snježana Križanić*, Katarina Tomičić-Pupek and Neven Vrček University of Zagreb, Faculty of organization and informatics Varaždin, Croatia DOI: 10.7906/indecs.23.5.4 Regular article Received: 30 April 2025. Accepted: 1 September 2025. ABSTRACT The diversity of students’ learning paths is crucial for acquiring knowledge. Although there are digital learning environments that provide many opportunities for managing the learning process, the rapid development of technologies can cause disruptions in the realisation of targeted engagement scenarios. Monitoring educational content use and increasing interaction frequency can contribute to better performance management and achievement of learning outcomes. Data and process mining methods and tools play a significant role in the research of performance and deviations. Anonymized real data from one elective university course was collected and processed to create a dataset for the application of clustering and decision tree analysis in the KNIME Analytics Platform and for creating a process model in a process mining tool. The results show behavioural patterns for three clusters and provide insight into interaction types by identifying variables related to content engagement as effective discriminators for student grouping. The process model illustrates the diversity of engagement in choosing learning paths through the course (based on 51 cases performing 52 distinct activities with an average of 233 activities), while retaining the focus on the assignment deliverables. Insights obtained from the analyses are useful for the effective implementation of digital learning environments illustrating that no exceptional scenarios occurred in the course in terms of deviations in behaviour with the digital learning platform in relation to similar teaching and learning paradigms provided by the same teachers and that more interactive features combined with new technologies would be useful in providing more personalized learning paths. KEY WORDS data mining, clustering, decision tree, process mining, educational data CLASSIFICATION ACM: H33 JEL: D83, M00, O31 *Corresponding author, : ; +385 42 390 893; *FOI, Pavlinska 2, HR – 42 000 Varaždin, Croatia S. Križanić, K. Tomičić-Pupek and N. Vrček INTRODUCTION Commitment to ensure engaged learning and provide feasible learning paths fitting various behavioural patterns is a desired feature while using digital learning environments. Teachers strive to develop a predictive model for improving teaching and course content management strategies based on real data. By using data on students’ interaction and performance, teachers can discover learning patterns, supporting nearly personalized learning paths for students with different behavioural types. Data and process mining methods and techniques promise to contribute to revealing patterns and sequences, creating visualizations and analytics offering to predict student performance and deviations, expected use-cases, enabling data-driven education management. The expected aim of a more efficient education management is to provide insights for all stakeholders in how to design and maintain a satisfying customer journey (i.e., a student journey through a course) consisting of enough (but not too much) challenging (but not too hard to pass) activities, through delivering interaction touchpoints and resources enabling the acquisition of desired learning outcomes. Recent developments in educational process mining have focused on discovering student learning paths and analyzing behavioral patterns within digital learning environments [1]. Educational data mining (EDM) and process mining have emerged as powerful analytical approaches for understanding and improving educational processes. The systematic application of data mining techniques in educational settings has been extensively documented, with researchers demonstrating the potential for analyzing student performance patterns and predicting academic outcomes [2]. The implementation of multimodal learning analytics has opened new avenues for understanding help-seeking behaviors and student interactions with both automated systems and human experts [3]. Comprehensive systematic reviews have confirmed that classification algorithms are most frequently applied in educational settings for evaluating student academic outcomes and identifying at-risk learners, while clustering techniques are commonly used for behavioural profiling and dropout prediction [4, 5]. These reviews emphasize that no single model uniquely predicts student performance, and the effectiveness of approaches depends heavily on data quality and contextual factors [5]. Although the motives for analysing student behaviour are often diverse, in this case the main goal was to investigate whether any exceptional scenarios occurred. This was done with consideration of the disruptive impact of new technologies on teaching and learning, and with a focus on what could be inferred from the measured engagement levels in one elective course at a higher education institution. The aim of this study is to explore student behavioural patterns in a digital learning environment using data and process mining techniques in order to identify engagement levels and detect potential deviations impacting course design and teaching strategies. The article is organized in four main sections as follows. The Literature Review section identifies previous achievements in the field of data and process mining in education. The Methodology section describes the research design and procedures employed in this study. In the Data subsection, the dataset used for the research is presented in detail. Subsequently, the subsection Clustering and Decision Tree Procedure explains the application of data mining techniques, specifically clustering and decision tree analysis. Similarly, the Process Mining subsection describes the process mining approach applied in this study. The results are discussed in the Results section. The article concludes with the Discussion and Conclusion. LITERATURE REVIEW The reviewed literature demonstrates a growing interest in applying data-driven approaches, particularly clustering, classification, and process mining in order to analyse and improve 468 Data and process mining in analysing student behaviour educational processes. Numerous studies employed process mining techniques such as process discovery and conformance checking to uncover students’ learning paths, behavioural patterns, or system-level process inefficiencies. Clustering methods, especially k-means and Expectation-Maximization, were frequently used to group students based on their interactions or performance levels, while classification algorithms like Decision Trees, Naïve Bayes, and Support Vector Machines were applied to predict academic outcom (...truncated)