Software Cost Estimation using Fuzzy Decision Trees

This paper addresses the issue of software cost estimation through fuzzy decision trees, aiming at acquiring accurate and reliable effort estimates for project resource allocation and control. Two algorithms, namely CHAID and CART, are applied on empirical software cost data recorded in the ISBSG repository. Approximately 1000 project data records are selected for analysis and experimentation, with fuzzy decision trees instances being generated and evaluated based on prediction accuracy. The set of association rules extracted is used for providing mean effort value ranges. The experimental results suggest that the proposed approach may provide accurate cost predictions in terms of effort. In addition, there is strong evidence that the fuzzy transformation of cost drivers contribute to enhancing the estimation process.