Dietary Patterns and Sample-Relative Nutritional Vulnerability among University Students and Lecturers in Bayelsa State, Nigeria: K-Means, Random Forest and Shapley Analysis

Background: This study examined dietary patterns and a constructed, sample-relative index of dietary vulnerability among 216 university students and lecturers surveyed in Bayelsa State, Nigeria, in 2026. A study-specific ten-food-group Dietary Diversity Score (DDS), total food-frequency intensity (TFFI) and nutrient-dense food-frequency score (NDFS) were combined using an equal-weighted, min-max normalised formula. K-means explored dietary profiles; Random Forest was used only to assess concordance with the index-defined upper-tertile NVI category; and mean absolute SHAP values described feature importance within that fitted classifier. From the results, Mean DDS was 9.63/10 (SD = 0.66), indicating a marked ceiling effect. The three-cluster solution had weak separation (silhouette = 0.133) and is interpreted as exploratory. Random Forest achieved 95.4% hold-out accuracy and ROC AUC of 0.988, but these values reflect reproduction of a rule derived from the same dietary variables, not independent prediction of clinical nutritional status. Dairy had the highest mean absolute SHAP importance; direction cannot be inferred from mean absolute values alone. The analysis illustrates exploratory profiling of this institutional sample.