An Impact of Poisoning Attacks on Machine Learning Algorithms

Machine learning models, ubiquitous in domains like natural language processing and image recognition, are vulnerable to adversarial poisoning attacks, where malicious actors manipulate training data to induce erroneous predictions. This study delves into the susceptibility of Support Vector Machine (SVM) models to such attacks, employing the Gradient Ascent Poisoning (GAP) approach. The findings reveal a substantial impact of GAP on SVM performance, leading to a notable decline in accuracy from 81% to 73%. Beyond presenting empirical evidence, the paper highlights the broader challenges of detecting and mitigating poisoning attacks, underscoring the need for robust security measures in machine learning systems. By illuminating the effectiveness of adversarial poisoning techniques and their ramifications, this research contributes to the ongoing endeavor to enhance the resilience of machine learning models in practical applications.