Interrupt Trace Fusion for Enhanced Website Fingerprinting Attacks under Defensive Mechanisms

Machine learning has significantly enhanced the effectiveness of website fingerprinting attacks, increasing the success rate of privacy leakage to 93.7%. Website fingerprinting attacks analyze various side-channel signals to deduce the specific websites a user visits, thus posing a threat to user privacy. In response to this threat, advanced defense mechanisms have been developed, such as randomized timers. These defenses alter the pattern of interrupt events, reducing the success rate of interruptbased side-channel attacks to 1.8%, thereby significantly disrupting attackers' ability to accurately infer information. To counter these effective defenses, we propose an interrupt trace fusionbased attack model. By combining interrupt traces from multiple attack sources, this model captures a more comprehensive set of interrupt features, thereby improving both the accuracy and stability of the attacks. Compared to single-source methods, this fusion model can still effectively extract useful information under strong defensive measures, enhancing overall attack performance. Experimental results show that this model raises the attack success rate from 93.7% to 94.9% under no defense. Under robust defenses, its accuracy increases from 1.8% to 45.3 %.