머신러닝 기반 수리조선 수요예측 모델 개발 및 특성 분석
Development and Feature Analysis of a Machine Learning-Based Ship Repair Demand Forecasting Model
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
This study aims to enhance the operational efficiency of shipyards by accurately forecasting the highly volatile demand in the ship repair market. For this purpose, a 5-year daily virtual dataset reflecting time-series characteristics was generated and refined through feature engineering. A comparative analysis was conducted on the performance of several prediction models, including Ridge, Random Forest, XGBoost, LightGBM, SARIMAX, and a Stacking Ensemble combining them. Hyperparameters were optimized using a ‘Broad to Narrow Search’ strategy. Furthermore, key influencing factors were identified through SHAP analysis, an Explainable AI (XAI) technique. The performance evaluation showed the SARIMAX, Stacking Ensemble, and Ridge models demonstrated the highest predictive accuracy, achieving R2 scores of 0.75 or higher. The SHAP analysis identified that daily port traffic, international oil prices, and shipping market conditions are the most critical factors in determining ship repair demand. Consequently, this study presents a systematic methodology for ship repair demand forecasting. The developed model and factor analysis can be utilized as crucial foundational data for ship repair yards to optimize resource allocation and establish data-driven operational strategies, thereby contributing to enhanced responsiveness and efficiency in a changing market.
Abstract
This study aims to enhance the operational efficiency of shipyards by accurately forecasting the highly volatile demand in the ship repair market. For this purpose, a 5-year daily virtual dataset reflecting time-series characteristics was generated and refined through feature engineering. A comparative analysis was conducted on the performance of several prediction models, including Ridge, Random Forest, XGBoost, LightGBM, SARIMAX, and a Stacking Ensemble combining them. Hyperparameters were optimized using a ‘Broad to Narrow Search’ strategy. Furthermore, key influencing factors were identified through SHAP analysis, an Explainable AI (XAI) technique. The performance evaluation showed the SARIMAX, Stacking Ensemble, and Ridge models demonstrated the highest predictive accuracy, achieving R2 scores of 0.75 or higher. The SHAP analysis identified that daily port traffic, international oil prices, and shipping market conditions are the most critical factors in determining ship repair demand. Consequently, this study presents a systematic methodology for ship repair demand forecasting. The developed model and factor analysis can be utilized as crucial foundational data for ship repair yards to optimize resource allocation and establish data-driven operational strategies, thereby contributing to enhanced responsiveness and efficiency in a changing market.