Android malware detection using Transformer, Decoder and Encoder models

With a rising amount of malware targeting Android, identification and analysis have become increasingly challenging. Malware is growing exponentially, making conventional ways of removing it more and more ineffectual. While previous authors have put forward a number of excellent methods. These days, malware employs incredibly complex and advanced methods. Therefore, static malware analysis alone is unable to identify it. We provide a novel basic network design called Transformer, Encoder, and Decoder models for the Android malware detection process, which are dependent purely on attention systems and without recurrence or convolutions. We collect two static characteristics from Android apps using Androguard: Application Programming Interface (API) calls and permissions. We train and test our methods using a dataset which is called malwaredataset. We demonstrate that the transformer, encoder, and decoder models extend effectively to different applications by accurately classifying malware and benign file elements using large and minimal training data. The research's results show that the suggested transformer, encoder, and decoder methodologies are successful in detecting malware and have average accuracy in categorizing.