A Methodological Framework for Time-Series Forecasting of Clinical Outcomes in Kenyan Smallholder Farming Systems

"background": "Smallholder farming systems are critical for food security and livelihoods, yet they face complex, interacting challenges affecting human and animal health. Predictive analytics for clinical outcomes in these settings are underdeveloped, lacking robust methodological frameworks that account for the temporal dynamics and multifactorial nature of disease drivers. ", "purpose and objectives": "This study aimed to develop and evaluate a novel methodological framework for forecasting clinical incidence in integrated smallholder systems. The primary objective was to test the predictive performance of a bespoke time-series model against observed clinical data. ", "methodology": "We constructed a longitudinal dataset from smallholder agro-pastoral households, integrating fortnightly clinical records for key livestock species with concurrent meteorological, management, and market data. The core forecasting model was a Seasonal Autoregressive Integrated Moving Average with eXogenous regressors (SARIMAX), specified as \ (B) \ (Bˢ) \ᵈ\D yt = \ (B) \ (Bˢ) \ + =1ᵏ \ x{i, t. Model fit was assessed using rolling-origin cross-validation, with uncertainty quantified via 95% prediction intervals. ", "findings": "The SARIMAX (1, 1, 1) (0, 1, 1, 26) model incorporating rainfall lag and herd vaccination coverage significantly outperformed a baseline ARIMA model, reducing the mean absolute scaled error by 32%. Forecasts for clinical mastitis incidence in dairy cattle showed high accuracy up to a 12-week horizon, with prediction intervals reliably capturing observed variance. ", "conclusion": "The proposed framework provides a statistically robust and operationally viable tool for anticipatory health management in smallholder contexts. It successfully integrates heterogeneous data streams to generate actionable forecasts. ", "recommendations": "Development practitioners and veterinary services should adopt similar forecasting approaches to enable proactive interventions. Further research should focus on integrating the model into digital decision-support platforms for frontline workers. ", "key words": "predictive modelling, One Health