Digital game-based math learning environments (math DGBLE) are promising platforms that provide students with opportunities to master conceptual understanding and cultivate mathematical thinking, on which the contemporary mathematics education places an emphasis. Literature on learning support in digital game-based learning (DGBL) rarely investigate learners' support-use behaviors and interaction patterns in relation to math learning. We addressed this research gap in this exploratory mixed-methods study. We designed and developed a packet of learning supports (i.e., Task Planner and Math Story) in a math DGBLE. Task Planner is designed to assist learners' systematic and planned efforts for math problem solving whereas Math Story features historical stories and real-life applications of math concepts. With the data collected via mixed-methods approach, we extracted six clusters of learning-support-use behaviors via unsupervised machine learning technique (i.e., Gaussian Mixture Model), including 1) skills development and application of mathematical problem decomposition, 2) conceptual knowledge development, 3) metacognitive mathematical connections, 4) metacognitive regulation, 5) information selection using cognitive aids, and 6) sustained motivation for necessary aversive practices. Qualitative multi-cases study revealed nuanced details regarding learners’ interactions with the learning supports in DGBLE. Results showed that the designed in-game learning supports facilitated individual meaningful and mindful math problem-solving experiences. The findings suggested that a mixed-methods research design integrating machine learning with multi-cases study could act as a tool for learner behavior and interaction pattern research that informs the design of adaptive and effective DGBL.