DEEP LEARNING MALWARE DETECTION FOR ANDROID MOBILE PHONES UTILIZING CCN AND TRANSFER LEARNING MODELS

Abstract

For the most recent generation of smart mobile devices, Android is the operating system that is most widely used. Numerous Android applications have been created for this operating system and have grown to be indispensable in our day-to-day activities. Unfortunately, a variety of Android malware has developed as a result of the steady stream of these apps. This malware has somehow infiltrated the system and gravely undermined system security protocols by installing additional packages, making API calls, and granting rights. To protect user privacy and prevent the greatest harm, it is crucial to identify and categorize Android malware. Numerous studies have already been conducted on the various methods for identifying and categorizing Android malware. This paper presents a deep learning strategy that is based on the concepts of convolutional neural networks and transfer learning. To identify and categorize malicious software intended to infect Android devices, this model uses a range of parameters, filter sizes, epoch counts, learning rates, and layers. This model was tested using the Drebin dataset, which contains 215 characteristics. The accuracy value of the CNN model is 99.09%. The F1-score, precision, and recall are the additional statistical values. Through innovative feature architecture and careful performance evaluation, the CNN model employs deep learning to identify Android malware, increasing accuracy and user protection. The model outperforms current methods in terms of accuracy.