Boosted Decision Trees for Project Risk Assessment and Pricing

In this paper, we present a predictive modeling approach to project risk assessment and pricing using modern machine learning techniques, namely boosting. In the first section, we present a broad overview of boosting fundamentals and techniques used to build predictive classifiers. We focus in particular on " Stumps" , which are one-level decision trees, and Alternating Decision Trees (ADT), which are more complex tree models. We also show how to use boosting techniques for multi-class prediction problems as well as for multi-class probability estimates. In section 2, we present the results of the experiments conducted with various boosting methods using project management data. The implications of using boosted decision trees on risk factor identification and prioritization is presented. We also show the main contribution of the prediction part to the estimation of project outcome probability and severity. In section 3, the estimated probabilities are fed into a Monte Carlo simulation algorithm which provides a better estimate of project financials. The impact of prior estimates on project performance, as estimated by the Monte Carlo process, is measured through a variation of the Kullback-Leibler divergence that we introduce for such purpose. The derived estimates can be used as an additional input to a decision support system for the profiling and the pricing of complex project portfolios.