Statistical modelling final portfolio

Mapa de Krigagem Ordinária de índices pluviométricos nos estados do MATOPIBA

This technical report develops a predictive model for beverage pH using historical manufacturing data from ABC Beverage to support quality control and regulatory compliance. We conducted exploratory data analysis on 2,571 production records, identifying key predictors including Mnf_Flow, Temperature, Pressure_Vacuum, and Carb_Rel as the strongest correlates with pH levels. After preprocessing (median imputation, column standardization, and categorical encoding), we trained and compared five machine learning models using 5-fold cross-validation: Multiple Linear Regression, Random Forest, XGBoost, K-Nearest Neighbors, and MARS. Random Forest emerged as the best performer with an RMSE of 0.10 and R-squared of 0.68, outperforming all other approaches due to its ability to capture nonlinear relationships between process variables. Feature importance analysis revealed that flow rates, temperature, and pressure settings have the greatest influence on final pH, providing actionable insights for manufacturing teams. The final model generated 267 predictions on the evaluation dataset, with predicted pH values falling within a realistic range of 8.17 to 8.79. We recommend deploying this model for real-time batch predictions, prioritizing monitoring of top-ranked variables, and retraining quarterly to maintain accuracy as production conditions evolve.

STAT 427, Chapter 8 Homework, Exponential smoothing models

ACTIVIDAD SEMANA 5