Android Malware Detection Method Based on CNN and DNN Bybrid Mechanism

With the continuous upgrading and development of malware attack methods, traditional detection methods have shown a series of serious problems such as low classification accuracy, easy overfitting, and high false alarm rate when facing new malware attacks. To address these challenges, this study introduces an innovative deep convolutional neural network (D-CNN) method that cleverly integrates permission features and API call graphs. Learn high-level abstract representation through DNN and combine it with CNN to build multiscale feature representation, aiming to improve the performance of the detection model and enhance the resistance to new malicious attacks. In order to ensure the standardization and representativeness of the data, the Min–Max method is first used to normalize the permission characteristics to ensure the standardization of the data. Second, the ant colony optimization method is used to achieve dimensionality reduction and prevent over-fitting. This article conducts experiments on Drebin and Google Play Store datasets. The results prove that the hybrid structure of D-CNN exhibits a deeper understanding of the data structure and achieves an accuracy of 96.80%, enabling more comprehensive and accurate malware detection and classification. It outperforms single deep learning methods in detection performance.