FORECASTING INDUSTRIAL PH LEVELS: COMPARATIVE STUDY OF SARIMA, REGRESSION TREES AND CONTROL CHART DIAGNOSTICS

Implementation of Statistical Process Control (SPC) techniques in food and beverage industry are crucial to deliver consumable product that meets customer expectations. This study investigated industrial pH forecasting and process stability in a syrup manufacturing facility. We analyzed 1,247 pH observations with three objectives: (1) Quantify instability via control charts, (2) Model pH dynamics using Seasonal Autoregressive Integrated Moving Average (SARIMA) and Classification And Regression Trees (CART), and (3) Develop diagnostic frameworks for unstable processes. Methodologically, Exponentially Weighted Moving Average (EWMA) charts assessed stability; Box-Cox transformed SARIMA (λ=2) with seasonal differencing was used for forecasting; CART identified variable importance. Control charts revealed profound instability: 83.3% of points violated 3σ limits; run tests significant (p<0.001). For SARIMA, (1,0,1)(0,1,1)₁₂ achieved significant parameters (p<0.0001) with improved residual diagnostics versus non-seasonal ARIMA, though minor autocorrelation remained at lag 12 (p=0.003). CART explained training R²=18.86% and test R²=17.93% of pH variation, identifying filling weight and sodium benzoate as key predictors. Crucially, this study demonstrates that forecasting retains diagnostic utility even in unstable environments: SARIMA residuals provide seasonal fingerprints of assignable causes, while CART thresholds guide intervention priorities. SARIMA(1,0,1)(0,1,1)₁₂ demonstrated superior residual properties: eliminated back forecast warnings (present in ARIMA), reduced autocorrelation at lag 24 (p=0.017 vs 0.040), and explicitly modeled 12-period seasonality. While process instability persists, SARIMA provides diagnostic fingerprints of assignable causes through seasonal parameters (SMA₁₂=0.9846, T=513.12) and residual patterns. We conclude that SARIMA offers enhanced short-term forecasting capability, but process intervention remains essential for reliability. The study advocates for integrated instability-informed forecasting combining SARIMA diagnostics, real-time control charts, and expanded sensor deployment.