Implementation of Six Sigma DMAIC and Random Forest for Defect Prediction in a Tire Manufacturing Laboratory

Abstract

The tire industry faces quality control challenges because undetected defects may affect user safety and export competitiveness. This study aims to integrate Six Sigma DMAIC and Random Forest to predict defects using laboratory testing data from PT Multistrada Arah Sarana Tbk. The dataset contains 15 quality characteristic attributes and one defect status target variable. Tire_ID was excluded because it serves as an identifier, resulting in 14 features. Class distribution shows a 3.27:1 ratio, with Non-Defect at 76.6% and Defect at 23.4%, indicating imbalance, which was addressed by adjusting the classification threshold to 0.45 without resampling. DMAIC was employed as the quality control framework, while Random Forest was applied during the Analyze and Improve phases. Testing results achieved 96.70% accuracy, 91.19% precision, 95.09% recall, 93.10% F1-Score, and 0.971 AUC-ROC. Feature importance identified Cure_Temperature (21.87%) and Uniformity_Index (17.54%) as dominant factors. The integrated approach increased recall by 20.79 percentage points compared with conventional methods, supporting earlier detection and effective corrective actions. It strengthens the transition from reactive to predictive quality control, prioritizing critical parameters and evidence-based continuous process improvement in laboratory operations.