Evaluating model performance and prediction accuracy of fuzzy logic and bayesian networks with business expert inputs

Abstract

This paper evaluates and compares two expert-in-the-loop decision-support paradigms Fuzzy Logic (FL) and Bayesian Networks (BN) for barangay-level business risk assessment using the BizLocator Analytics dataset of Cauayan City, Philippines. Local governments need transparent, data-driven tools that can operate under uncertainty, sparse data, and evolving economic conditions. FL and BN are both well-established approaches for modeling uncertainty, yet they are rarely examined side by side on the same dataset with the same expert knowledge. To address this gap, the study develops parallel FL and BN models grounded in identical features and informed by the same pool of business and experts. The FL model uses expert-defined triangular and trapezoidal membership functions, together with a compact set of IF–THEN rules that encode linguistic concepts such as “Low Compliance,” “Vulnerable Barangay,” and “High Risk.” The BN model encodes expert-elicited causal relationships as a directed acyclic graph and learns conditional probability tables from data under Dirichlet priors. A unified preprocessing pipeline is applied, and nested stratified cross-validation is used to avoid optimistic bias and to support paired statistical tests. Both models are evaluated on discrimination (Accuracy, F1-score, ROC–AUC, PR–AUC) and probabilistic quality (Brier score, Expected Calibration Error, reliability diagrams). Results show that BN achieves slightly higher discrimination and notably better calibration, while FL offers superior case-level interpretability through rule and membership visualizations. Expert validation confirms that most BN edges are causally plausible and FL rules covers the majority of decisions. The findings suggest that a hybrid deployment using BN as the calibrated scoring backbone and FL as an explanation and policy-communication layer can provide accurate, transparent, and actionable decision support for local business risk governance and long-term planning. Overall, the study demonstrates how expert-guided artificial intelligence can strengthen evidence-based regulation while preserving human oversight and accountability in practice across diverse barangay.