A very early estimation of software development time and effort using neural networks

In spite of years of research and development, formal structured estimation of time and effort required to develop a Management Information System (MIS) is still an open problem. Usual estimation techniques applied by now are supported by the not so realistic premise of requirements stability, and often human experts are required to apply them. This paper considers models of estimation based on metrics available on early design phase. Our research work aims to develop formal estimation models for time and effort needed for MIS development. These models use development team efficiency, requirements volatility, development speed and system complexity as input parameters. We also identify which input metrics are adequate for measuring system’s cognitive complexity and found that useful metrics can be obtained automatically from the system users´ data views very early on the life cycle with independence of the technology used and without human intervention. We tested the metrics estimation capability using Artificial Neural Networks (ANN), and thus confirmed an existing functional relation among input and output metrics (time and effort). Once trained, the ANN predicts effort needed with a 15% average error and time needed with a 30% average error. Eje: I - Workshop de Ingeniería de Software y Base de Datos Red de Universidades con Carreras en Informática (RedUNCI)