Improving DGA-Based Malicious Domain Classifiers for Malware Defense with Adversarial Machine Learning