Email phishing remains a critical cybersecurity threat due to uncertainty in content features such as linguistic ambiguity, overlapping patterns, and noisy labels, which challenge conventional evaluation methods. Existing uncertainty-handling approaches, including fuzzy and probabilistic models, typically represent ambiguity using a single membership dimension or probability distribution, limiting their ability to simultaneously model truth, indeterminacy, and falsity as independent components. To address this limitation, this study proposes a neutrosophic entropy-based Multi-Attribute Decision Making (MADM) framework for the objective selection of phishing detection algorithms under uncertain conditions. The proposed framework models algorithm performance using Single-Valued Neutrosophic Sets (SVNS), capturing tri-component uncertainty (truth, indeterminacy, and falsity) independently. A neutrosophic entropy measure is employed to derive objective criteria weights based on the intrinsic dispersion of performance information, reducing reliance on subjective expert weighting. The weighted evaluations are aggregated using Neutrosophic Weighted Averaging (NWA) and Neutrosophic Weighted Geometric (NWG) operators to obtain final rankings. The framework is applied to five supervised learning algorithms Random Forest, Decision Tree, Logistic Regression, Multilayer Perceptron, and Support Vector Machine evaluated on a balanced dataset of 50,000 emails using precision, recall, accuracy, training time, and evaluation time as criteria. The results demonstrate that the neutrosophic entropy-based approach provides an uncertainty-aware and structurally objective decision support mechanism for algorithm selection in phishing detection systems.
