Keywords: Software metrics, Bayesian belief networks, decision tree, nearest neighbor, risk prediction.
In order to find some relationships between the risk of the software and the metrics extracted from the code of the national project related software system, in this paper, we applied four classical intelligent classification methods which were most used in data mining fields, namely Bayesian belief networks (BBN) [1, 2, 3], nearest neighbor (NN)  rough set (RS) and decision tree (DT) , , to predict the risk of the core source code of a spectrum processing software system which is a "863" National Project related to the Large Sky Area Multi-Object Spectroscopic Telescope (LAMOST) Project in China.
In Section 3, we will present the results of our experiments in applying Bayesian belief networks, nearest neighbor, rough set and decision tree models to risk prediction.
Firstly, we employed Bayesian belief networks, which was a statistical classification method.
1] used Bayesian belief networks for risk prediction and found that it had many advantages comparing with traditional methods which were regression-based.
Table II presents the results of using Bayesian belief networks for individual metrics in two categories.
Table III shows us the results of using Bayesian belief networks for individual metric in three categories.
Topics include customized benchmark generation using MDA, predicting change impact in architecture design with Bayesian belief networks
, the changing role of architects, and a declarative approach to architectural reflection.