Inquiry-based learning with interactive simulations is widely used in science education. However, learners often struggle to engage productively and benefit fully without appropriate support. To provide such support, it is crucial to better understand the inquiry patterns learners exhibit and how these relate to learning outcomes across different activities. In this study, we apply a clustering approach to investigate learners' interactions with STEM interactive simulations through three distinct paradigms of feature abstraction. For each paradigm, we construct a transition matrix per learner and simulation, capturing the sequence of inquiry actions. These matrices are then processed using a spectral clustering pipeline to identify groups of learners with similar behavioral patterns. We apply this method to log data from 135 learners engaging with two different simulations. Across both simulations, we consistently identify two distinct clusters. These clusters show similar behavioral patterns and are associated with comparable learning outcomes, with the larger cluster typically demonstrating a deeper understanding of the underlying scientific concepts. Moreover, we find that most learners maintain the same inquiry pattern across simulations, suggesting that the identified behaviors reflect stable approaches to inquiry rather than task-specific strategies. Our results also reveal that prior knowledge and activity order influence the strategy learners adopt. Finally, we show that while the three paradigms yield largely overlapping clusters, each highlights different aspects of learner interaction, demonstrating the relative robustness of our findings to the chosen data representation.