A comparison of AdaBoost, Bootstrap ANN and SVM project performance prediction models in the early planning stage

• Main Authors: CHEN CHIEN HAO, 陳建豪
• Other Authors: WANG YU REN
• Format: Others
• Language: zh-TW
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/51616758031103650608

碩士 === 國立高雄應用科技大學 === 土木工程與防災科技研究所 === 99 === Before the commencement of a construction project, feasibility analyses such as technical, economical and environmental feasibility analyses must be conducted and these are parts of the early planning. This research intends to investigate the relationship between early planning practice and final project outcomes. A project definition tool, Project Definition Rating Index ( PDRI ) , is incorporated in this research as a survey instrument for the building construction industry in Taiwan. The PDRI evaluation scores are taken as the input variable and project success ( cost and schedule ) are set as the output variable. Artificial Intelligence ( AI ) models ( ANNs Bagging, Adaboost, and support Vector Machines ) are developed and validated using data collected from 105 building construction project in Taiwan. Among the 105 sample projects, 75 projects are randomly chosen as the training data and the remaining 30 projects are use as testing data. Prediction models are developed to predict project outcomes (project cost or schedule success ). Using ANNs Bagging techniques, a total of 11 classifiers are developed and the overall cost success prediction accuracy rates is 63% for this particular model. For ANNs Adaboost classification models, the overall cost success prediction accuracy rate is 67% after weighted voting. For LSSVM cost success prediction model, the final overall accuracy rate achieved is 80%. The schedule success prediction outcomes are 63%, 70% and 70% overall accuracy rate for the ANNs Bagging Adaboost and LSSVM models respectively. In summary, the LSSVM prediction models produce the best prediction results for both cost and schedule success predictions. The results can provide the owner, contractors and projects managers, valuable information to evaluate early planning and predict potential project performance.