기계학습 알고리즘을 통한 제조원가 추정분석 연구

[Purpose]The purpose of this study is to apply a machine learning model for estimating the components of manufacturing costs, including material, labor, and factory overhead costs, and to identify the contribution of predictive variables using SHAP values. [Methodology]This study analyzed 110,556 firm－year observations from manufacturing firms disclosing cost statements between 1990 and 2022. Target variables were material, labor, and factory overhead costs, while predictive variables included 26 accounting and 6 non－accounting factors. Machine learning techniques like XGBoost, Random Forest, LightGBM, and MLP were applied to assess estimation performance. Ensemble methods enhanced accuracy, and SHAP analysis evaluated predictive variable contributions. [Findings]The application of machine learning algorithms demonstrated superior performance in predicting material and labor costs, with XGBoosting, GBM, and blending ensemble methods exhibiting high estimation accuracy and stable error control. [Implications]This study demonstrates the applicability of machine learning for estimating costs that are difficult to predict using traditional regression methods reliant on strict statistical assumptions. The refined models also offer academic value, enabling their use in future policy－related studies based on cost analysis data.