Ioc-Mldroid: a Hybrid Approach for Detection of Android Based Malicious Applications Using Indicators of Compromise (Iocs) and Machine Learning

Android operating system has taken over the mobile ecosystem across the world and thus is a prime target of more advanced malware that uses its ability to obfuscate, dynamically load code, encrypted communications and even environment aware evasion behaviors. These properties severely circumscribe the ability of conventional signature-based detection systems, and are problematic to either all-static or all-dynamic analysis. Although machine learning-based detection methods have promise, most current solutions are black-box systems with no forensic transparency [5][6] , and are not as applicable in security operations and digital investigations. This paper presents an android malware detection system being a hybrid between the static-dynamic framework and powered by Indicators of Compromise (IoCs) and machine learning. The framework uses a constrained dual-VM setup consisting of an analysis host based on Ubuntu and an instrumented android emulator. Standard analysis tools are used to extract the static IoCs, which consist of permissions, manifest components, use of sensitive APIs, cryptographic artifacts, certificate metadata, and embedded network indicators. Controlled application execution can be used to obtain dynamic IoCs by monitoring and instrumenting runtime behaviors related to behavioural indicators like network communication, filesystem activity and sensitive API invocation. A framework IoC correlation and normalization engine is used to combine heterogeneous non-uniform indicators which are static and dynamic, into structured feature vectors. A random Forest model that has been chosen due to its strong ability to predict high-dimensional data, as well as the ability to have an interpretable component, are used to classify these vectors. In addition to binary classification, the framework produces a structured forensic intelligence report, which records the IoCs behind each detection decision, which in turn facilitates evidence-based and explainable malware detection. The suggested framework puts the emphasis on the detectability, interpretability and forensic relevance. It improves resilience to evasion by obfuscation, but does not increase decision-making transparency through the correlation of fixating and moving IoCs. The work provides an expandable and explainable android malware detection architecture that can be used in academic research, enterprise security operations, and advanced digital forensic investigation, and which may be extended in the future to the scale of validation and on-device detection.