A proof-of-concept hybrid knowledge-based decision support system for precision livestock nutrition: translating a small-sample silage ramie substitution trial in Xiangxi Yellow Cattle into computable evidence

Introduction Precision livestock farming requires decision support systems (DSS) that can translate biologically rich but small feeding-trial datasets into transparent and actionable nutritional evidence. This study developed a proof-of-concept hybrid knowledge-based DSS using a controlled silage ramie substitution trial in Xiangxi Yellow Cattle. Methods Twenty-one cattle were allocated to five dietary treatments. The system integrated an experiment-derived evidence-rule engine, a 21-record case library for case-based reasoning (CBR), and ensemble machine-learning predictors based on Random Forest and Gradient Boosting. Module outputs were combined through a confidence-aware meta-reasoning layer. Ramie versus control contrasts were adjusted using the Benjamini-Hochberg false-discovery-rate procedure, and machine-learning performance was internally assessed using leave-one-out cross-validation. Results Under the tested conditions, complete substitution of silage corn with silage ramie was associated with improved serum antioxidant and immune biomarkers and lower lipid peroxidation: CAT +15.4% (P = 0.0005; q = 0.0011), SOD +174.5% (P = 0.0002; q = 0.0011), T-AOC +14.7% (P = 0.0008; q = 0.0014), MDA -51.5% (P = 0.0065; q = 0.0097), IgA +45.1% (P = 0.0005; q = 0.0011), and IgG +41.8% (P = 0.0005; q = 0.0011). Carcass-yield indicators declined modestly. Internal cross-validation suggested that ensemble learning captured diet-sensitive serum biomarkers better than complex production traits (SOD: R 2 = 0.791; CAT: R² = 0.545; net meat rate: R² = -0.263). Discussion The proposed framework provides a transparent evidence-translation workflow for converting small-sample feeding-trial results into computable, case-specific decision-support assets. However, the system should be interpreted as a proof-of-concept prototype rather than an externally validated feeding recommendation; larger multi-batch validation and economic modeling are required before farm-level deployment.