Runtime Malware Detection using hardware features

The low-level microarchitectural malware prediction methods have shown good results in finding the viruses than the earlier software-based techniques. But these methods have failed to make an impact on the false detection rates which affect the precision of the detection scheme largely. When a person installs an application from a third-party website like filehippo.com or download.com, these methods tend to ignore significant attributes of an application like checksum, signature of the app creator that in turn makes an impact on the precision of the prediction methods due to more false prediction rates. In order to achieve a lesser false alarm rates we track the impact made on the hardware features by the applications and train the basic ML classifiers and boost their efficiency using ensemble methods and represent a clear performance optimization of the ML classifiers using minimal number of HPCs, thereby, minimizing the required number of runs for which we make use of few ML based learning models and two ensemble learners for labelling the applications and provide precision in terms of precision, efficiency & hardware overhead. We have enhanced the precision and efficiency by reducing the false prediction rates of the basic methods by using ensemble techniques. We have reached the target of more than 98%accuracy from previous 85%. It is almost impossible to define any specific classifier that can give the best prediction.