International Research Journal of Education and Technology

Current AI analytics excels at pattern recognition but struggles with reasoning under uncertainty and novel scenarios. This paper presents Probabilistic Neurosymbolic Analytics (PNA), a hybrid framework merging belief functions for uncertainty quantification with neurosymbolic reasoning for logical inference. We demonstrate the framework through a healthcare case study on sepsis prediction, showing how PNA maintains 91% accuracy while providing human-interpretable explanations. Comparative analysis against pure neural networks reveals PNA's superior robustness to distributional shifts (15% accuracy improvement) and explainability (78% user task success vs. 45% baseline). These results validate the viability of uncertainty-aware, explainable smart analytics for real-world decision systems.

Index Terms—Smart data analytics, neurosymbolic AI, belief functions, uncertainty reasoning, explainable AI, creative problem solving.