A learning adaptation cases technique for Fuzzy Analogy-based software development effort estimation

The aim of this paper is to enhance the Fuzzy Analogy technique for software effort development estimation. Fuzzy Analogy selects the similar projects that will be used in the adaptation step according to the definition of the qualification `closely similar´. The adopted definition consider two projects as closely similar if their similarity is in the vicinity of 1. The qualification `closely similar´ is represented by a fuzzy set defined by a fixed threshold which is obtained experimentally from the environment. However, in many cases the available empirical knowledge may not allow estimators to fit the adequate fuzzy representation of the qualification `closely similar´. In this study, we propose an approach to learn this fuzzy representation from the similarities obtained in the retrieval step of the Fuzzy Analogy technique. The proposed method provides for each new project, an adequate threshold by using the quasi-arithmetic mean operators. Indeed, the quasi-arithmetic means operators use weighted similarities to calculate the threshold that often ensures the selection of the closest projects in the adaptation step. This paper also presents an empirical validation of the proposed approach based on the COCOMO´81 dataset.