HYBRID ARTIFICIAL INTELLIGENCE MODELS COMBINING NEURAL NETWORKS AND SYMBOLIC REASONING FOR COMPLEX PROBLEM SOLVING
Abstract
Hybrid artificial intelligence models combining neural networks and symbolic reasoning emerged as a powerful paradigm for solving complex computational problems that require both data-driven learning and logical inference. This study evaluated the performance of hybrid neuro-symbolic AI systems compared with standalone neural network and symbolic reasoning models using benchmark datasets. A quantitative experimental design was applied with a sample size of 300 test instances drawn from three standard datasets used in machine learning and reasoning tasks. Performance was assessed using accuracy, precision, recall, F1-score, interpretability, error rate, and generalization measures. The hybrid model achieved the highest accuracy (93.8%), precision (92.7%), recall (92.9%), and F1-score (92.8%), outperforming neural networks and symbolic systems. Error rate remained lowest at 6.2%, while generalization reached 94.1%, indicating strong adaptability to unseen data. Results demonstrated that integration of neural learning with symbolic reasoning improved predictive performance, logical consistency, and robustness under noisy conditions. Neural networks performed well in learning patterns but showed limited interpretability, while symbolic systems demonstrated strong interpretability but weaker adaptability. The hybrid model maintained a balance between both capabilities, improving explainability without reducing accuracy. Findings confirmed that neuro-symbolic integration provided a scalable solution for complex decision-making environments requiring transparency and reliability. The study contributed to advancing explainable artificial intelligence by demonstrating the effectiveness of hybrid architectures in bridging learning and reasoning paradigms for real-world applications.
Keyword-Artificial Intelligence, Explainable AI, Hybrid Systems, Neural Networks, Neuro-Symbolic Reasoning, Symbolic Logic
