Evolutionary Antivirus Signature Optimization

This work presents an automatic methodology able to improve ma-chine-generated signatures for Android Malware detection. The technique relies on a population-less evolutionary algorithm and uses an unorthodox fitness function that incorporates unsystematic human experts knowledge in the form of a set of rules of thumb. The proposed optimization algorithm does not require to rank the individuals, as exploiting experts knowledge, the resulting population of candidate solutions is not a totally ordered set any more. Experimental results show that the resulting signatures are of good quality and more accurate than the original ones, lowering both false positives and negatives.