Validating game-based learning assessment of students’ computational thinking practices using Bayesian networks and machine-learning based detectors

Zoombinis is an award-winning digital game aligned with four computational thinking practices—problem decomposition, pattern recognition, abstraction, and algorithm design. The study manipulates two levels of scoring: Evidence identification (EI)—identifying outcomes from a single puzzle-level,—and evidence accumulation (EA)—summarizing observed outcomes across puzzles and levels. For three puzzles, machine learning detectors provide information about players’ computational thinking practices. For the remaining 45 puzzle-levels, only basic information (e.g. degree of player success) is available. These are combined with two EA algorithms: the average of detector outputs, and a Bayesian network. This study uses data from 716 students complete pre- and post-assessments. The correlation among the measures of computational thinking was positive yet modest, suggesting that the measures focus on different aspects of computational thinking.