INTEGRATING REINFORCEMENT LEARNING WITH EXPLAINABLE AI FOR REAL-TIME DECISION MAKING IN DYNAMIC ENVIRONMENTS

Abstract

Artificial Intelligence (AI) has significantly transformed autonomous decision-making by enabling intelligent agents to learn optimal strategies through continuous interaction with dynamic environments. Among various machine learning paradigms, Reinforcement Learning (RL) has emerged as one of the most effective approaches for sequential decision-making, allowing agents to maximize cumulative rewards without requiring labelled training data. RL has demonstrated remarkable success in applications such as autonomous driving, robotics, industrial automation, healthcare, cybersecurity, and smart resource management. However, despite its outstanding performance, most deep reinforcement learning models function as black-box systems whose internal decision-making processes remain difficult for human users to interpret. This lack of transparency reduces user confidence, limits system accountability, and creates challenges in safety-critical domains where understanding AI behaviour is essential. To overcome these limitations, Explainable Artificial Intelligence (XAI) has been introduced to improve the interpretability and transparency of intelligent systems. This research proposes an integrated Explainable Reinforcement Learning framework that combines the Proximal Policy Optimization (PPO) algorithm with SHapley Additive exPlanations (SHAP) to provide real-time, human-understandable explanations for autonomous decisions. The proposed platform is implemented using FastAPI, Stable-Baselines3, SHAP, SQLite, HTML, CSS, JavaScript, and Chart.js to create a web-based interactive environment for agent training, performance monitoring, decision analysis, and visualization. The framework supports multiple benchmark environments, including CartPole-v1, MountainCar-v0, and Acrobot-v1, enabling comprehensive evaluation under different dynamic scenarios. Experimental analysis demonstrates that the proposed system achieves stable learning performance while simultaneously generating accurate feature attribution and transparent explanations for every decision. Interactive dashboards, audit logging, and real-time monitoring further improve system usability, accountability, and trustworthiness. The proposed Explainable Reinforcement Learning framework provides a scalable foundation for deploying transparent AI systems in healthcare, intelligent transportation, robotics, finance, cybersecurity, and industrial automation, thereby facilitating the responsible adoption of autonomous decision-support technologies in real-world environments.