臺灣大學: 土木工程學研究所曾惠斌陳明Tran, MinhMinhTran2013-04-012018-07-092013-04-012018-07-092011http://ntur.lib.ntu.edu.tw//handle/246246/255680Construction industry plays a major part in any nation economy. However, the construction industry tends to face high risk due to the particular characteristic of the environment and high competition. Therefore, many researches have been conducted to find an appropriate model to forecast bankruptcy in construction sector. Artificial Neural Network (ANN) using Back Propagation Algorithm has been applied in this area since the early 1990s, and has been showed the promising outcome. Accordingly, in this study Back Propagation Network (BPN) was selected to construct a model in bankruptcy prediction for construction industry. In the previous study employing ANN methods, the sample-matching technique was usually used, which lead to sample selection biases, likely due to ANN’s inability to tackle between-class imbalance problem. In this research Back Propagation Network (BPN) using over-sampling techniques with all available firm-year data was proposed so as to tackle between-class imbalance challenge. The two over-sampling techniques used were: Enforce training and Synthetic Minority Over-Sampling TEchnique (SMOTE). The empirical result of this study showed that the BPN using SMOTE was out performed the BPN original and EBPN. Accordingly, BPN using SMOTE are suggested as an alternative to the existing model5889365 bytesapplication/pdfen-US違約概率預測建築業人工神經網絡反向傳播 算法類間的不平衡強迫訓練合成少數股東採樣技術default probability predictionconstruction industryArtificial Neural NetworkBack Propagation Algorithmbetween-class imbalanceEnforced trainingSynthetic Minority Over-Sampling TEchniquethesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/255680/1/ntu-100-R98521746-1.pdf