A Time-Series Forecasting Model for Yield Improvement in Ugandan Manufacturing Systems: A Methodological Evaluation

"background": "Manufacturing systems in developing economies often lack robust, data-driven methodologies for continuous yield improvement. Existing forecasting approaches are frequently ill-suited to the high-variability, low-data-density environments typical of such settings, leading to suboptimal production planning and resource allocation. ", "purpose and objectives": "This article presents a methodological evaluation of a novel time-series forecasting model designed specifically to measure and predict yield improvements in manufacturing systems. The objective is to provide a formalised, adaptable framework for plant engineers to enhance production efficiency through more accurate yield projections. ", "methodology": "The proposed methodology integrates a Seasonal AutoRegressive Integrated Moving Average (SARIMA) model with exogenous variables (SARIMAX) to account for operational factors. The core model is defined as \ (B) \ (Bˢ) \ ablaᵈ\ ablasD yt = \ + \ (B) \ (Bˢ) \ + =1ᵏ \ x{i, t, where yt is the yield series and xi, t are exogenous regressors. Model parameters were estimated using maximum likelihood, with inference based on robust standard errors to mitigate heteroscedasticity. ", "findings": "The methodological evaluation, applied to a case study, demonstrated that the integrated SARIMAX model reduced one-step-ahead forecast error by approximately 18% compared to a standard ARIMA benchmark. Diagnostic checks confirmed model adequacy, with residual autocorrelation plots showing no significant structure. ", "conclusion": "The evaluated methodology provides a statistically sound and operationally relevant framework for yield forecasting in resource-constrained manufacturing environments. It offers a substantial improvement over conventional, less adaptive time-series models. ", "recommendations": "Manufacturing plant engineers should adopt this integrated modelling approach, incorporating both temporal patterns and contextual operational data. Further research should focus on automating model selection and integrating real-time data streams for dynamic updating. ", "key words": "Time-series forecasting,