MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202611070233 A) filed by Ims Engineering College, Ghaziabad on June 04, 2026, for A Method For Detecting Bias In Machine Learning Models Using Explainable Artificial Intelligence.
Inventors include Dr. Ramesh Kumar Verma; Mr. Vivek Jain; Mr. Basudeo Singh Roohani; and Mr. Awdhesh Kumar.
The application for the patent was published on August 07, 2026, under issue no. 32/2026.
Abstract: The present invention relates to a computer-implemented method for detecting bias in machine learning models using explainable artificial intelligence. The method receives a machine learning model, input dataset, prediction outputs, and fairness configuration data including protected attributes and fairness thresholds. Explanation data is generated for input instances using explainable artificial intelligence techniques to determine feature contribution values associated with model outputs. Prediction distributions and explanation distributions are compared across protected groups, reference groups, and intersectional subgroups to identify outcome disparity and explanation disparity. The method further detects proxy features by analysing association with protected attributes and material influence on model predictions. Counterfactual bias testing is performed by modifying protected or proxy attributes under predefined constraints and observing changes in model outputs. A bias severity score is computed based on disparity, proxy influence, counterfactual instability, and error imbalance. An explainable bias report is generated identifying affected groups, contributing features, severity scores, and mitigation recommendations. Accompanied Drawing [FIGS. 1-2]
Disclaimer: Curated by HT Syndication.
