SecureDML:An Intelligent Framework for Preventing Poisoning Attacks in Distributed Machine Learning Systems

The security and protection of models in dis-tributed machine learning (ML) systems require high emphasis on adversarial threats, including poisoning attacks. This study contains a complete framework that integrates different advanced techniques to monitor poison attacks and prevent such attacks for the effective functioning of machine learning systems. The proposed system integrates hybrid encryption for security, and a subsequent anomaly detection method using autoencoders. SHapley Additive exPlanations-based interpretability method is used to enhance model transparency. Hybrid encryption combines the RSA and AES methods to keep data and model parameters secret, and autoencoders provide effective identification of poisoning attack patterns through abnormal data observations. This method is implemented using multimodal datasets such as CIFAR 100 and AG News datasets. Finally,the effectiveness of this method can be evaluated using confusion matrix, comparison graphs. It works as a comprehensive solution that benefits various ML applications, such as healthcare, autonomous vehicles, Large Language Models, etc., for enhancing security along with integrity protection.

@article{T2026,
  title     = {SecureDML:An Intelligent Framework for Preventing Poisoning Attacks in Distributed Machine Learning Systems},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {17},
  number    = {2},
  year      = {2026},
  publisher = {The Science and Information Organization},
  author    = {Archa A. T and Kartheeban K},
  doi       = {10.14569/IJACSA.2026.0170295},
  url       = {https://doi.org/10.14569/IJACSA.2026.0170295}
}