Expert systems and knowledge representation in AI-driven decision-making: a comprehensive survey

This survey provides a comprehensive overview of expert systems and methods of knowledge representation in AI-based decision-making in various areas of application. We review the evolution of knowledge-based systems that are advancing from systems that are rule-based to the contemporary hybrid forms that are symbolically reasoned, mixed with neural learning. The survey analyzes the current techniques of knowledge representation, which include ontologies, Knowledge Graphs, semantic networks, Bayesian networks, and reasoning processes. We conduct the review of our applications in four major contexts, i.e., healthcare (clinical decision support, diagnostic systems), finance (fraud detection, credit scoring), education (intelligent tutoring systems), and smart cities (traffic management, energy optimization). In a structured comparative analysis, we discovered that hybrid AI solutions are effective in various areas that combine the interpretability of symbolic reasoning and neural learning ability. Bottlenecks of knowledge acquisition, scalability problems, issues associated with trust, and systems integration complications have been noted as some of the major issues investigated in the survey. A summary of the research directions in the future includes Neural-Symbolic integration, explainable AI, lifelong learning, and unified knowledge systems. The work can help researchers and practitioners with a systematic presentation of state-of-the-art approaches, methodologies, and open issues in knowledge-based AI systems.