Game-Based Learning Prediction Model Construction

Game-based learning (GBL) is increasingly recognized as an effective tool for teaching diverse skills, particularly in science education, due to its interactive, engaging, and motivational qualities, along with timely assessments and intelligent feedback. However, more empirical studies are needed to facilitate its wider application in school curricula. A significant challenge is designing and implementing valid in-game assessments crucial for measuring student progress and providing reliable references for instructors’ intervention decisions. Stealth assessment, guided by the evidence-centred design (ECD) framework, offers a promising solution but requires more specific guidelines for full effectiveness. In this study, we present a granular, framework-supported pipeline to systematically implement stealth assessments in a GBL environment. This pipeline involves constructing an ECD framework, generating features, selecting appropriate models, preprocessing data, evaluating model performance, and conducting model inference on a black-box computational model. We validate the effectiveness of this pipeline by assessing the performance of these computational models and identifying distinct behavioural patterns between high and low performers. Our analysis highlights potential areas for improvement in the design of stealth assessments within digital games for learning. Furthermore, we discuss the generalizability of the proposed pipeline and outline limitations for future research to address.