Study of data-mining-based project construction duration prediction model

• Main Authors: CHI-MIN LIU, 劉奇敏
• Other Authors: Leu, Sou-Sen
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/09036172935791737974

博士 === 國立臺灣科技大學 === 營建工程系 === 103 === This research collects completed construction projects of petrochemical industry in Taiwan as the database for analysis on construction project duration. With the initial compilation founding of nearly 25% of the projects have time extension records, though lower than the 40% found in other relative researches, the result shows that current scheduling factors cannot fully predict feasible construction project duration. The research employs projects from this database with valuation approach, by using direct and indirect factors to develop a predictable construction scheduling pattern, which can provide effective information for reviewing the feasible construction scheduling of new investments, and to help avoid project delays which may affect investment profits. A Top-down approach is used throughout the research for factor analysis; using Principle Component Analysis (PCA) method of data mining in restructuring factors and derived factors from the projects, in order to obtain Principal Components scores (PCs) which are then combined with Back Propagation Neural Network (BPNN) to produce a duration prediction mechanism, becoming a process of project period prediction called PCA-BPNN model. The results are then verified by 10 new projects, with outcomes showing root mean squared errors (RMSE) from 0.05 to 2.01, fulfilling demands in practice by predicting a feasible construction period and enhance project management.