Computational Intelligence in Empirical Software Engineering

The objective of Empirical Software Engineering is to improve the software development and maintenance processes and consequently the quality of theirs various deliverables. This can be achieved by evaluating, controlling and predicting some important attributes of software projects such as development effort, software reliability, and programmers productivity. One of the most interesting sub-field of ESE is software estimation models. Software estimation models are used to predict some critical attributes of some entities that are not yet exist. For example, we often need to predict how much a development project will cost, or how much time and effort will be needed., so that we can allocate the appropriate resources to the project. In general, estimation models relate the attribute to be predicted to some other attributes, that we can measure now, by using mathematical formulas or other techniques such as neural networks, case-based reasoning, regression trees and rule-based induction. Currently, our research concerns software cost estimation models. We have developed an innovative approach referred to as Fuzzy Analogy for software cost estimation. Nevertheless, this approach can be used to evaluate and predict other attributes such as reliability, quality, safety, and maintainability. In this paper, we present some results of our recent research related to the cost estimation field. 1. Cost estimation: Techniques and challenges Accurate and timely prediction of the development effort and schedule required to build and/or maintain a software system is one of the most critical activities in managing software projects, and has come to be known as ‘Software Cost Estimation’. In order to achieve accurate cost estimates and minimize misleading (underand over-estimates) predictions, several cost estimation techniques have been developed and validated. These techniques may be grouped into two major categories: § parametric models, and § non-parametric models. Parametric models are derived from statistical or numerical analysis of historical projects data (simple/multiple/stepwise regression, Bayesian approach, polynomial interpolation, ...)