Preprocessing and Feature Engineering of Gameplay Logs for Adaptive Mathematics Learning Dataset Construction

Abstract

This study reports the preprocessing and feature engineering of gameplay logs collected from a first-grade elementary mathematics educational game prototype. The study aimed to transform raw gameplay activity records into structured analytical datasets that can support early performance description and serve as an initial data preparation stage for future adaptive mathematics learning research. A limited trial was conducted with nine first-grade students who played six sequential game levels covering early numeracy topics, including counting, number ordering, number reading, place value, and mixed review. The gameplay logs captured event-based student interactions, including session identity, level, question, selected answer, correctness status, attempt count, help usage, response time, score, and event type. The data processing workflow included data validation, cleaning, anonymization, data type handling, event filtering, feature engineering, dataset aggregation, and descriptive analysis. The preprocessing stage produced 262 clean gameplay log records consisting of 171 answer events, 28 help events, 54 level completion events, and nine game completion events. Feature engineering generated analytical indicators such as level accuracy, average response time, total help usage, average attempt count, performance category, student state, and initial adaptation action. The final outputs were organized into answer-level, level-level, student-level, and adaptive feature datasets. The anonymized dataset and data dictionary are provided as supplementary materials to support reproducibility and future reuse. The results indicate that raw gameplay logs can be converted into structured datasets for early learning analytics and preliminary adaptive learning data preparation, without making claims about learning effectiveness or final adaptive model performance.