Anonymized Moodle interaction dataset for learning style prediction (FSLSM) and educational data mining

This repository contains an anonymized educational dataset extracted from a Moodle Learning Management System (LMS), supporting research on Learning Analytics, Educational Data Mining (EDM), and the automatic identification of learning styles based on the Felder-Silverman Learning Style Model (FSLSM). The data was collected from postgraduate courses involving pedagogical intervention and includes student interaction logs, assessment performance, and forum participation metrics.Dataset ContentsThe repository consists of CSV files representing different dimensions of student engagement: acesso_recursos.csv (Resource Access Logs): Detailed logs of student interactions with learning resources. Key features: user_hash (anonymized ID), resource_type (e.g., quiz, url, forum), access_timestamp , duration , and action_type . forum_participacao_anon.csv (Forum Participation): Quantitative data on social interactions within course forums. Key features: Number of posts created, replies posted, and total interactions per student. df_features.csv (Aggregated Feature Set): The processed dataset used for machine learning modeling (Clustering and Stacking Ensemble). Includes aggregated metrics such as: total_time : Total time spent on the platform. quiz_attempts : Number of attempts on assessments. avg_grade : Average performance score. review_gain : Performance improvement between attempts (re-evaluation gain). interaction_diversity : Count of unique resource types accessed. Anonymization & Ethics To ensure compliance with data protection regulations (LGPD/GDPR) and ethical research guidelines: Direct Identifiers: All direct identifiers (names, emails, student IDs) have been removed. Hashing: Student identities were replaced with irreversible SHA-256 hash tokens ( user_hash ) to maintain data integrity for longitudinal analysis without compromising privacy. Data Minimization: Only fields relevant to the pedagogical analysis were retained. Methodology This dataset supports a study that applies K-Means Clustering to identify behavioral profiles and a Stacking Ensemble Classifier (Random Forest + XGBoost) to predict learning styles. The data processing pipeline involved feature engineering, timestamp parsing, and handling missing values (imputation with 0 for sparsity). Potential Uses Replication of Learning Style prediction models. Analysis of student engagement patterns in LMS. Development of adaptive recommendation systems. Benchmarking of clustering algorithms in educational contexts.