A data-driven decision-support framework for predictive quality management in FMCG manufacturing

Abstract Quality management in high-volume manufacturing increasingly requires data-driven approaches that complement conventional retrospective analysis. This study developed a retrospective decision-support framework for FMCG biscuit manufacturing by integrating Six Sigma DMAIC with regression modelling, time-series forecasting, Failure Mode and Effects Analysis (FMEA), and a proposed Statistical Process Control (SPC) monitoring layer. The analysis used archived production records and a 372-record modelling dataset from PT XYZ, an Indonesian biscuit manufacturer. Packing and process related defects accounted for approximately 91.5% of the category-based defect weight. Under chronological regression validation, Linear and Ridge Regression outperformed Support Vector Regression; however, all three models performed poorly when only one-day antecedent explanatory information was used, limiting the prospective interpretation of the strong contemporaneous results. Among ARIMA, Prophet, Exponential Smoothing, and two simple benchmark forecasts, ARIMA achieved the lowest RMSE, MAE, and MASE on the retained 30-record chronological test set. FMEA prioritized conveyor stoppage, coding-machine-related defects, and uneven oven temperature, while SPC remained a proposed monitoring component. The incremental contribution of the framework lies in integrating retrospective association analysis, temporal forecasting, risk prioritization, and prospective monitoring within a single DMAIC-based decision-support sequence while distinguishing the evidential role of each component. Therefore, the findings support the framework as a retrospective, case-based decision-support approach rather than a validated real-time predictive-control system. Further prospective evaluation using clearly antecedent predictors, repeated temporal validation, and operational SPC implementation is needed before predictive-control effectiveness can be established. Graphical Abstract