The purpose of this case study is to demonstrate how to utilize machine learning approaches to analyze student process data for validating and informing digital game-based assessments (DGBAs) with an evidence-centered game design (ECgD). The first analysis was conducted to examine whether students' mastery of the overall skill required by the game can be well predicted by task-related behavioral features and if the selected key features map onto the evidence model of the ECgD. Specifically, we extracted 27 behavioral features as the indicators of students' gameplay activities from the evidence trace files and modelled them using a machine learning algorithm--support vector machine with recursive feature elimination--to identify the key features for prediction. The key features were in turn used to predict students' mastery of the overall skill. Results showed that students' retry attempts on two assessment tasks were found to be most influential for prediction with a moderate to high training and testing accuracy. The second analysis was conducted to examine whether the number of learning opportunities is sufficient for evaluating students' mastery of the overall skill as well as determine the optimal number of learning opportunities for evaluation. The approach of long short-term memory networks was used to model students' time-series behavioral features across multiple learning opportunities for predicting their acquisition of the overall skill. Results suggested that five learning opportunities were a good balance between evaluation accuracy and practical feasibility, and they were sufficient for evaluating students' mastery of the overall skill given the DGGA tasks.