Detecting Android malware in an effective way requires more than just looking at individual behavioral features, as the previous studies did in the last few years - we also need to understand the relationships between the Android application components. In this research study, we introduce a knowledge-graph integrated framework that combines system-call interactions and inter-app bindings into a combined knowledge network. We apply graph embedding techniques to this structure and integrate the results with established behavioral features, our approach reveals latent patterns that frequency-based methods typically miss. In the real-world experiment on a large, real-world dataset of multiple Android applications, the our model showed substantial improvements over baseline classifiers. These results emphasize the value of relational insights and draw a future research direction toward more resilient, graph-aware mobile security solutions.