국가종합전자조달시스템 복수예비가격에서 다중선형회귀 및 정규화 모델 기반의 예정가격 예측

초록

Korea Online E-Procurement System (KONEPS) is a developing integrated procurement system that processes 64 trillion won in transactions, annually. As of 2021, roughly 668,964 procurement companies were competing for successful bids. A standard price was set to prevent bidding at unreasonably high or low prices. The multiple preliminary method determines the standard price by randomly selecting 15 preliminary prices, with four prices set at random lottery, to ensure transparency in competitive bidding, as knowing the standard price would give a company unfair advantage. This paper discusses the Pearson’s correlation coefficient in predicting the standard price, using a multiple linear regression model. A regularized regression model solved the multicollinearity problem, finding predictive power as a major indicator. All four models showed high explanatory power with over 99% of R-square, with an error margin of 1%. The elastic net model best demonstrated both explanatory power and evaluation index. The ridge regression model proved most effective with independent variables, and the Lasso regression model showed variable selection characteristics. The RMSE index indicated greater error in the Lasso model, while the ridge model showed greater error in the MAPE index. All four models are expected to increase competitiveness in predicting the standard price.

키워드