Prediction Model of Software Development Effort Based on Input Output Correlation

Abstract

Predicting software development effort accurately is crucial for the timely delivery of quality-assured products within a reasonable timeframe. Underestimating and overestimating effort will lead to serious consequences. This paper uses Input Output Correlation method to sort the importance of a variety of attributes which have impacts in predicting software development effort, and uses RBF neural network to train these attributes in order to find some attributes which accurately predict effort. It could remove redundant and irrelevant attributes effectively. In order to find internal rules between associated attributes and effort, this paper constructs decision tree to extract internal rules. Finally, it uses Matlab 6.5 to do simulation experiments. The result shows that this method effectively improves prediction accuracy.