Software Effort Estimation Using Data Mining Techniques

Abstract

This paper describes an empirical study undertaken to investigate the quantitative aspects of application of data mining techniques to build models for Software effort estimation. The techniques chosen are Multi linear regression, Logistic regression and CART.Empirical evaluation using three fold cross validation procedure has been carried out using three bench marking datasets of software projects, namely, Nasa93, Cocomo81, and Bailey Basili. We observed that: (1) CART technique is suitable for Nasa93 and Nasa93_5. (2). Multiple Linear Regression is suitable for Nasa93_2, Cocomo81s, Cocomo81o and Basili Bailey. (3). Logistic Regression is suitable for Nasa93_1, Cocomo81 and Cocomo81e. It is concluded that data mining techniques tend to help estimating in the best way possible as they are objective and are applicable to unlimited sets of data.