Using regression analysis , Kalman filter , and neural networks to the prediction of tendering price on roadway construction

• Main Authors: Pei-chu Hsu, 徐培哲
• Other Authors: nono
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/10126315070326085395

碩士 === 逢甲大學 === 土木工程所 === 95 === Although domestic public infrastructure budget is notified during tendering (internet notice in accordance with Article 27 of Procurement Act), bottom price used to select vendor is setup by the proprietor. In addition, due to fierce market competitions, most vendors make deductions based on estimated price during tendering to increase the likelihood of winning a tender. Based on existing public engineering tender procedures in this country, most infrastructure project vendors are selected using the lowest bid mechanism. However, this type of selection method is likely to result to inappropriate tender strategies. Excessive bidding result to vicious price cut among bidders. It will in turn affect the quality of tender. Therefore, this study has constructed more precise engineering bidding price prediction model based on related mathematical statistics basics to assist vendors increase the likelihood of winning bid and optimize resource distribution and application. This study adopts engineering tenders of road infrastructure sponsor in 2005 as study subjects. Public invitation for tender in the amount of NT$50million or lower and NT$1million or higher (not exceeding threshold amount value) serves as the engineering tender sample. Based on engineering tender related data such as bottom price, price of award, budget price, contract period (calendar day), and bid bond, lowest price of award and proprietor ceiling price prediction model is constructed using Regression Analysis, Kalman Filter, and Artificial Neural Networks. In consideration of regional influential factors, tender data source is divided as to region into northern region, central region, southern region, eastern region, and integrated. Results of comparative analysis conducted show that the error range for these three prediction methods is between 10.63％ and 24.29％. Meanwhile, the respective prediction methods do not show absolute systematic advantage or disadvantage. Moreover, prediction on bottom price is more feasible. As for prediction for price of award, since more influential factors are involved and there are more unpredictable decision-making behaviors during tendering, it is impossible to construct a set of more reasonable quantified mathematical prediction model.