Bolstering Stealth Assessment in Serious Games

摘要

Stealth assessment is an unobtrusive assessment methodology in serious games that use digital player traces to make inferences of players' expertise level over competencies. Although various proofs of stealth assessment's validity have been reported, its application is a complex, laborious, and time-consuming process. To bolster the applicability of stealth assessment in serious games; a generic stealth assessment tool (GSAT) has been proposed, which uses machine learning techniques to reason over competence constructs, player log data and assess player performance. Current study provides empirical validation of GSAT by applying it to a real-world game, the abcdeSIM game, which was designed to train medical care workers to act effectively medical emergency situations. GSAT demonstrated, while relying on a Gaussian Naive Bayes Network, to be highly robust and reliable, achieving a three-level assessment accuracy of 96%, as compared with a reference score model defined by experts. By this result, this study contributes to the alleviation of stealth assessment's applicability issues and hence promotes its wider uptake by the serious game community.