We propose a probabilistic neural network (PNN) approach for simultaneously estimating values of software development parameter (either software size or software effort) and probability that the actual value of the parameter will be less than its estimated value. Using real-world software engineering datasets and V-fold sampling, we compare the PNN approach with the chi-squared automatic interaction detection (CHAID) approach and find that the PNN approach performs similar to the CHAID, but provides superior probability estimates. We also show how the method of odds likelihood ratios can be used to combine the PNN forecasted values with subjective managerial beliefs to improve probability estimates.
All Science Journal Classification (ASJC) codes
- Computer Science Applications
- Artificial Intelligence