Android code vulnerabilities early detection using AI-powered ACVED plugin

During Android application development, ensuring adequate security is a crucial and intricate aspect. However, many applications are released without adequate security measures due to the lack of vulnerability identification and code verification at the initial development stages. To address this issue, machine learning models can be employed to automate the process of detecting vulnerabilities in the code. However, such models are inadequate for real-time Android code vulnerability mitigation. In this research, an open-source AI-powered plugin named Android Code Vulnerabilities Early Detection (ACVED) was developed using the LVDAndro dataset. Utilising Android source code vulnerabilities, the dataset is categorised based on Common Weakness Enumeration (CWE). The ACVED plugin, featuring an ensemble learning model, is implemented in the backend to accurately and efficiently detect both source code vulnerabilities and their respective CWE categories, with a 95% accuracy rate. The model also leverages explainable AI techniques to provide source code vulnerability prediction probabilities for each word. When integrated with Android Studio, the ACVED plugin can provide developers with the vulnerability status of their current source code line in real-time, assisting them in mitigating vulnerabilities. The plugin, model, and scripts can be found on GitHub, and it receives regular updates with new training data from the LVDAndro dataset, enabling the detection of novel vulnerabilities recently added to CWE