On the Resilience of Shallow Machine Learning Classification in Image-based Malware Detection