Abstract: In the software development process, accurate estimation of software effort is of great significance for software project and the enterprise. In order to overcome the difficulties that there isn't an accurate number for new software project at early stages according to the attributes of history project dataset and the existing software effort estimation models can't deal with the fuzzy number effectively, case-based reasoning and generalized fuzzy number are integrated, and a case-based reasoning (CBR) model based on generalized fuzzy number is proposed for software effort estimation in this paper. The traditional similarity measure such as Euclidean distance is replaced by a new similarity measure based on generalized fuzzy number in CBR model. Furthermore, the fuzzy c-means clustering is applied to fuzz the accurate number in history project dataset. Moreover, particle swarm optimization (PSO) is employed to further optimize attribute weights of the model. Finally Desharnais data is adopted to examine the validity of the model. It is shown that the proposed generalized fuzzy numbers CBR model can improve the estimation accuracy in comparison with the commonly used Euclidean distance CBR. In addition, it is also shown that the model with optimized weight from PSO can improve the estimation accuracy.

Key words: software effort estimation, case based reasoning, generalized fuzzy number, weight optimization