Paper Details

Name: Parveen Kumar Goyal & Garima Tyagi
Country: India

Year: 2025
Volume: Volume-12, Issue-2 (July-December)
Page Number: 360-376
DOI: https://doi.org/10.5281/zenodo.18013678

ABSTRACT
This work investigates the urgent need for models that are simultaneously robust and responsible in Data Privacy Threat detection. In this paper a Hybrid Machine Learning Framework is to be used that fuses Convolutional Neural Network for feature learning, also an XGBoost classifier to be implemented which has been carefully optimized using optuna bayesian approach. It resulted in better classification performance with an AUC 0.999 and F1-Score of 0.9886, clearly outperforming unoptimized baselines. Importantly, conceptualization is taken beyond mere performance, such as by introducing a way to measure model interpretability and ethical reasoning. Hear a new comparative study quantifying important operational statistics such as and the Traceability Index to steer deploy resource efficiency, leverage Natural Language Processing to investigate and verify model explanations against the ethical compliance benchmarks. This paper devising a novel metricthat use the feature of textual model outputs, i.e., Inference Privacy Score to measure the privacy leakage risk and guarantees solution being not only high-performing but fully traceable and responsible.

Keywords: Hybrid Machine Learning Framework, Data Privacy Threat Detection, Optuna Bayesian Optimization, XGBoost Classifier, Inference Privacy Score, Ethical Compliance Audit.