Malware detection in Android environment using deep recurrent neural network and genetic algorithm

Android malware is a significant threat to the security and privacy of mobile devices, and its detection is of great importance in ensuring the integrity of user data and system performance. In the methods carried out in the past, most of the methods have low accuracy, so in order to face this challenge, in this thesis, the focus was first on pre-processing the data so that they are prepared for the next analysis. Various pre-processing techniques were used to clean and remove the duplicate data of the dataset and ensure the integrity and improvement of the data. After that, “genetic algorithm “and mutual entropy were used to select the most relevant features for malware detection. The “genetic algorithm “used its evolutionary search mechanism to discover different subsets of features, while cross-entropy was used to evaluate the cost and investigate the differences between these features. Through this process, a subset of ۷۲ features were identified as the most appropriate features for malware classification. With these selected features obtained, the next step involved training an LSTM network for Android malware classification. LSTM networks are particularly effective at capturing long-term dependencies and sequential patterns, making them well-suited for this task. The proposed approach was implemented and tested on Tuandromd Android malware dataset obtained from UCI site. The experimental results showed the effectiveness of the proposed method in identifying Android malware, so that we could achieve ۹۹.۵% accuracy, which was ۱% more optimal than the multifaceted deep neural network method.

کلیدواژه ها:

Android malware detection ، deep recurrent neural network ، genetic algorithm