Jasmine: A New Active Learning Approach to Combat Cybercrime

Abstract

Jasmine is a novel hybrid Active Learning method for Network Intrusion Detection• Jasmine queries uncertain, anomalous and randomly selected observations• Jasmine can dynamically adjust the balance between these query types• Jasmine can learn the best strategy during the labeling process• Jasmine is the first Active Learning method with dynamic updating • Jasmine performs better than existing static Active Learning methods