Scientific Reports (Sep 2024)
Prediction of android ransomware with deep learning model using hybrid cryptography
Abstract
Abstract In recent times, the number of malware on Android mobile phones has been growing, and a new kind of malware is Android ransomware. This research aims to address the emerging concerns about Android ransomware in the mobile sector. Previous studies highlight that the number of new Android ransomware is increasing annually, which poses a huge threat to the privacy of mobile phone users for sensitive data. Various existing techniques are active to detect ransomware and secure the data in the mobile cloud. However, these approaches lack accuracy and detection performance with insecure storage. To resolve this and enhance the security level, the proposed model is presented. This manuscript provides both recognition algorithms based on the deep learning model and secured storage of detected data in the cloud with a secret key to safeguard sensitive user information using the hybrid cryptographic model. Initially, the input APK files and data are preprocessed to extract features. The collection of optimal features is carried out using the Squirrel search optimization process. After that, the Deep Learning-based model, adaptive deep saliency The AlexNet classifier is presented to detect and classify data as malicious or normal. The detected data, which is not malicious, is stored on a cloud server. For secured storage of data in the cloud, a hybrid cryptographic model such as hybrid homomorphic Elliptic Curve Cryptography and Blowfish is employed, which includes key computation and key generation processes. The cryptographic scheme includes encryption and decryption of data, after which the application response is found to attain a decrypted result upon user request. The performance is carried out for both the Deep Learning-based model and the hybrid cryptography-based security model, and the results obtained are 99.89% accuracy in detecting malware compared with traditional models. The effectiveness of the proposed system over other models such as GNN is 94.76%, CNN is 95.76%, and Random Forest is 96%.
Keywords