POISONED DATA DETECTION USING FOR DISTRIBUTED MACHINE LEARNING TECHNIQUES

Distributed Machine Learning (DML) facilitates large-scale model training across multiple nodes, improvingefficiency and scalability. However, it is highly susceptible to data poisoning attacks, where adversaries injectmanipulated data to degrade model performance. Detecting poisoned data in DML is essential for ensuring modelreliability and security. This paper explores detection techniques such as anomaly detection, trust-based evaluation,and statistical analysis to identify malicious contributions. We implement and evaluate these methods in various DMLenvironments to assess their effectiveness in mitigating adversarial threats. Experimental results demonstrate that ourapproach significantly improves model robustness while maintaining computational efficiency.