Classification of public elementary students' game play patterns in a digital game-based learning system with pedagogical agent

This study investigates gameplay attributes that were used to classify student performance in a digital game-based learning system to determine if it will contribute to achieving learning gain. The study was conducted in selected public elementary schools which comprised of 10% of all grade four students in each school visited. Word Infection Version 4, a local-area-network digital game-based learning (DGBL) system with a pedagogical agent, and a pretest and posttest module which served as the tool to collect gameplay logs of students were developed. Also, a dashboard tool was developed to manage, facilitate and administer the game in a distributed network. Usability test results showed strong agreement on its usability, aesthetics and usefulness. Log attributes, gameplay patterns, and performance of elementary students' vocabulary learning were recorded then described using K-means algorithm to determine the different clusters of students' gameplay patterns and performance while using the system. Four clusters were produced to represent the different gameplay styles of the students: gaming, proficient, productive and idle. A model that classified game play patterns of student's performance using Naïve Bayes and J48 algorithms was produced. The accuracy and kappa statistic of the produced models were determined. Higher ratings in accuracy and kappa statistic were yielded by the decision tree algorithm; 52.34% and 0.216 respectively in comparison 42.88% and 0.062 respectively from Naïve Bayes.