Cross-course Process Mining of Student Clickstream Data -- Aggregation and Group Comparison

TL;DR

This study presents a data engineering and process mining pipeline to analyze Moodle clickstream data across multiple courses. By extracting event logs and final scores via SQL, transforming and standardizing course-section labels with KNIME/PM4KNIME, and exporting cross-course XES logs, the authors enable intra- and cross-course analysis of student learning paths using Disco and ProM. Key findings show that higher-performing students engage more with LMS activities and move more dynamically between sections, with stronger signals when data are aggregated at the section level and analyzed across courses. The work provides methodological contributions for cross-course LMS analytics and empirical guidance for optimizing course structures to facilitate navigable and repeating engagement patterns across sections and courses.

Abstract

This paper introduces novel methods for preparing and analyzing student interaction data extracted from course management systems like Moodle to facilitate process mining, like the creation of graphs that show the process flow. Such graphs can get very complex as Moodle courses can contain hundreds of different activities, which makes it difficult to compare the paths of different student cohorts. Moreover, existing research often confines its focus to individual courses, overlooking potential patterns that may transcend course boundaries. Our research addresses these challenges by implementing an automated dataflow that directly queries data from the Moodle database via SQL, offering the flexibility of filtering on individual courses if needed. In addition to analyzing individual Moodle activities, we explore patterns at an aggregated course section level. Furthermore, we present a method for standardizing section labels across courses, facilitating cross-course analysis to uncover broader usage patterns. Our findings reveal, among other insights, that higher-performing students demonstrate a propensity to engage more frequently with available activities and exhibit more dynamic movement between objects. While these patterns are discernible when analyzing individual course activity-events, they become more pronounced when aggregated to the section level and analyzed across multiple courses.

Figures (10)

• Figure 1: Data Preparation Process.
• Figure 2: Mapping from log data fields to XES parameters.
• Figure 3: Procedural Languages without filtering.
• Figure 4: Detail of Procedural Languages filtered.
• Figure 5: Data Warehouse & Business Intelligence path comparison.