ESTIMAÇÃO DE MÉTRICAS DE DESENVOLVIMENTO AUXILIADA POR REDES NEURAIS ARTIFICIAIS.

Abstract

Several modeling approaches for the process of development in software engineering able of subsidizing decision making in the management of project are being searched. Metric of softwares, process modeling and estimation techniques have been independently considered either taking into consideration the intrinsic characteristic of softwares or their constructive process. This research proposes a complete, simple and efficient model for representing the whole process of development which, based on a set of features of the process and basic attributes of softwares, yields good estimation metrics (time and effort) of the development of the software still at the beginning of the process. The model relates constructive characteristics of the process to each type of organization, for identifying classes of homogeneous behavior based on Kohonen Neural Network. Directly, from this classification, according to the basic attributes of each software being developed, metrics may be estimated supported by Feedforward Neural Networks. A prototype is specified in Unified Model Language (UML) and implemented to estimate metrics for the development of softwares. Comparisons of the obtained results with those available in literature are presented.