Performance comparison of local versus cloud malware detection on Android using machine learning techniques

Malware detection is a key component of mobile system security today. While standard virus detection software is an important tool in the fight against mobile intrusion, it cannot detect zero-day attacks due to its reliance on previously detected signatures. Recognizing first time attacks requires a different approach and commonly involves machine learning. The problem is that such approaches are more computationally intensive and can create a user experience problem if the mobile device is not responsive enough. An experiment was conducted using machine learning, specifically a genetic algorithm, comparing performance of zero-day attack analysis running on a mobile device versus using cloud resources. It was demonstrated that even with a high-end mobile device, the use of a hybrid model that uses a cloud-based server for running the machine learning module vastly out-performs the all-local model.