Addressing Noise and Efficiency Issues in Graph-Based Machine Learning Models From the Perspective of Adversarial Attack