SOFTWARE EFFORT PREDICTION: AN EMPIRICAL EVALUATION OF METHODS TO TREAT MISSING VALUES WITH RAPIDMINER registered

Missing values is a common problem in the data analysis in all areas, being software engineering not an exception. Particularly, missing data is a widespread phenomenon observed during the elaboration of effort prediction models (EPMs) required for budget, time and functionalities planning. Current work presents the results of a study carried out on a Portuguese medium-sized software development organization in order to obtain a formal method for EPMs elicitation in development processes. This study focuses on the performance evaluation of several regression-based EPMs induced from data after applying three different methods to treat missing values. Results show that regression imputation offers substantial improvements over traditional techniques (case deletion and mean substitution). All the machine learning methods were implemented in RapidMiner registered , one of the leading open-source data mining applications.