Feature Importance Analysis of SMV Gap and Manpower Variables on Garment Production Output based on Ensemble Learning

Abstract

In the labor-intensive garment manufacturing industry, there is a conventional assumption that achieving production targets largely depends on increasing the number of workers (manpower), causing evaluations of operational time efficiency to often be overlooked. To examine this assumption, this study aims to identify the dominant factors affecting garment production output using a Machine Learning approach based on Ensemble Learning methods, namely Random Forest and Gradient Boosting. The dataset consisted of 700 observations collected at 20-minute intervals, including variables such as actual Standard Minute Value (SMV), SMV gap, actual manpower, and manpower gap. The evaluation results indicate that the Random Forest model outperformed Gradient Boosting, achieving a Mean Absolute Error (MAE) of 4.55, Root Mean Square Error (RMSE) of 6.85, Mean Absolute Percentage Error (MAPE) of 19.12%, and an R² value of 0.758. In comparison, Gradient Boosting obtained an MAE of 4.88, RMSE of 7.21, MAPE of 20.78%, and an R² value of 0.733. Based on the best-performing model, the feature importance analysis revealed that actual SMV was the most dominant factor (>0.70), followed by the SMV gap (>0.20). In contrast, manpower variables had a very limited influence (<0.05). Unlike previous studies that generally focused on optimizing the number of workers, the novelty of this research lies in the simultaneous use of SMV gap and manpower gap variables within a predictive model. This approach provides empirical evidence that job complexity and operational time efficiency are significantly more critical in determining garment production output than merely increasing the workforce.