Research On Project Risk Evaluation And Prediction Based On Statistical Learning Theory

Because of the increasing risk factors, traditional risk management methods will not adapt to and satisfy the development of modern project management, and it is necessary to study the new theories and methods that related to project risk management. Statistical learning theory (SLT) which is based on the structural risk minimization provides a new idea to the research of intelligent and scientific project risk management. This dissertation researches on project risk management based on the SLT, and some intelligent methods of risk evaluation, early warning and prediction are proposed. It is not only significant for theory study, but also helpful for risk management practice.The choice of feature indicators is important for exact project risk evaluation and early warning. Distance evaluation technique (DET) is introduced to feature indicators extraction in the dissertation. Redundant evaluating indicators which are unrelated or negative to risk evaluation can be extracted and the most sensitive evaluating indicators can be found by DET automatically. The results of the application show that the veracity and efficiency of assessment and early warning will be improved by DET, and it is also available for decreasing the workload of data acquisition and saving project cost.The essential of project risk evaluation is pattern classification. Traditional pattern classification methods, such as neural network (NN) is based on the empirical risk minimization, and is concerned with the machine learning principles under the infinite-sample situation. They are not suitable for project risk assessment which is a typical small-sample problem. Support vector machine (SVM) which is developed in the framework of SLT is a new machine learning algorithm for small-sample problem. A novel intelligent evaluating model based on DET and SVM is presented in this dissertation, and satisfied results are obtained in its application.The high risk data of significant projects is infrequent, and they are more difficult to be evaluated by traditional methods. A one-class classification method called support vector data description (SVDD) is studied, and an intelligent early warning method based on DET and SVDD is also proposed. With this model the risk level can be distinguished only by one-class data of low risk projects. The results of its application show that the method is valuable for early warning with the shortage of high risk data.The time sequence prediction of project risk factors is another key problem of risk management. In most cases, traditional methods can't obtain the satisfied forecasting results because the time sequences are usually non-stationary and non-linear. A hybrid intelligent forecasting method based on empirical mode decomposition and support vector regression (SVR) is proposed. The example of project material price forecasting reveals that the hybrid model is better than single SVR both in one-step and multi-step prediction.