This investigation recommends the implementation of an optimal ensemble learning framework to improve the precision and durability of Android malware detection. The framework combines decision trees, deep neural networks, and support vector machines to detect a variety of malware patterns, leveraging the complementary capabilities of multiple machine learning classifiers. Among other static and dynamic aspects, we employ feature extraction techniques to examine the permissions, API calls, and behavioral traits of Android apps in order to conduct an analysis. In order to enhance classification performance and minimize false positives, a weighted voting mechanism is implemented to amalgamate the predictions of individual models. The experimental results indicate that the proposed ensemble method achieves superior detection rates, more precise results, and greater scalability in comparison to the more conventional single-model approaches. The framework improves mobile security in environments where threats are constantly changing by effectively supporting real-time malware detection.