Predicting project success using ANN-ensemble classificaiton models

The construction projects have become larger in scale and more complex in system engineering in the past few decades. The researchers and industry practitioners have recognized its potential impact to final project outcomes and started to put more emphasis on early planning process. Nevertheless, the early planning practice varies significantly throughout the industry. This research intends to benchmark the early planning practice for the building construction industry in Taiwan and develop ANN-ensemble classification models to predict project success using survey data. A benchmarking tool, Project Definition Rating Index (PDRI), is incorporated in the questionnaire survey to measure the status of early project planning. With the PDRI evaluation as the independent variable and project success (measured by cost and schedule performances) as the dependent variable, logistic regression as well as ANN-ensemble methods (Bootstrap Aggregation and AdaBoost) are adopted to develop prediction models. Information from a total of 92 building projects is collected for the analysis in this study. The analysis results indicate that the status of early planning is an important factor to final project success. The results also show that AdaBoost ANN model yields the best predictions when comparing to other models. In summary, it is shown that surveyed projects with better early planning are more likely to achieve project goals at completion and the ANN ensemble models can be used to predict project success. The research results provide valuable information for the researchers to further investigate the topic of early planning and for the industry practitioners to perform better early planning.