An Intellectual Zero Trust Security Framework Using Deep Reinforcement Learning for Predictive Threat Mitigation in AI-Based Fraud Detection Systems

Project Code :TCMAPY2461

Objective

The primary objective of this research is to design and implement a robust Intellectual Zero Trust Security Framework that integrates Hybrid_ZT_MA_DRL_lgm_Model and Meta-DQN for accurate fraud detection. The framework aims to leverage multi-agent reinforcement learning with ensemble modeling to improve predictive performance while handling imbalanced datasets. It focuses on comprehensive feature engineering, label encoding, and top-feature selection to optimize learning. Meta-DQN is implemented to enhance adaptive decision-making through experience replay and reward-driven learning. The system includes modular components such as Home, Register, Login, Classification, and Logout to ensure scalability and usability. Performance is evaluated using metrics including accuracy, precision, recall, and F1-score, demonstrating improvement over conventional models.

Abstract

Fraud detection in financial transactions remains a critical challenge due to evolving attack patterns and highly imbalanced datasets. This research introduces an Intellectual Zero Trust Security Framework employing two advanced models: the Hybrid_ZT_MA_DRL_lgm_Model and Meta-DQN. The Hybrid model integrates multi-agent Deep Reinforcement Learning with LightGBM to analyze transaction, behavioral, and authentication features independently while combining outputs through a weighted hybrid decision mechanism. The Meta-DQN model applies reinforcement learning with experience replay and adaptive reward functions to optimize decision-making against dynamic fraud patterns. Strong feature engineering, top-feature selection, label encoding, and SMOTE-based balancing improve predictive performance. Modular design includes Home, Register, Login, Classification, and Logout components, supporting scalable deployment. Evaluation on a fully balanced validation set demonstrates superior accuracy, precision, recall, F1-score, and AUC-ROC compared to conventional systems. This framework enhances proactive threat mitigation, intelligent decision-making, and robust fraud detection in AI-driven financial environments.

Keywords: Zero Trust, Deep Reinforcement Learning, Multi-Agent, Hybrid Model, Meta-DQN, LightGBM, Fraud Detection, Feature Engineering, SMOTE, Predictive Analytics

NOTE: Without the concern of our team, please don't submit to the college. This Abstract varies based on student requirements.