Maintainability Prediction: A Regression Analysis of Measures of Evolving Systems

In order to build predictors of the maintainability of evolving software, we first need a means for measuring maintainability as well as a training set of software modules for which the actual maintainability is known. This paper describes our success at building such a predictor. Numerous candidate measures for maintainability were examined, including a new compound measure. Two datasets were evaluated and used to build a maintainability predictor. The resulting model, Maintainability Prediction Model (MainPredMo), was validated against three held-out datasets. We found that the model possesses predictive accuracy of 83% (accurately predicts the maintainability of 83% of the modules). A variant of MainPredMo, also with accuracy of 83%, is offered for interested researchers.

• Type: Conference
• Year: 2005
• Where: ICSM
• Authors: Jane Huffman Hayes, Liming Zhao
• Keywords: Actual Maintainability, ICSM 2005, Maintainability, Maintainability Predictor, Software Engineering