A probabilistic model for predicting software development effort

Abstract

We use the naive Bayes model to forecast software effort A causal model is developed from the literature, and a procedure to learn Bayesian prior and conditional probabilities is provided. Using a data set of 40 real-life software projects we test our model. Our results indicate that the probabilistic forecasting models allow managers to estimate joint probability distribution over different software effort estimates. A software project manager may use the joint probability distribution to develop a cumulative probability distribution, which in turn may help the manager estimate the uncertainty that the project effort may be greater than the estimated effort. © Springer-Verlag Berlin Heidelberg 2003.