Research And Application Of Analogy To Early Software Estimation

In the early stage of software life circle, if the software requirements could not been precisely understood, and the size, effort, and duration of software been reasonably estimated, then the fine planning and execution to a project can not been implemented, and the software products can not been delivered on time, within budget, and to an agreed level of quality. Therefore the early estimation is considered to be one of the most important factors which contributed to the success of a software project. But, because almost all the early estimations were made under the basis of uncertain or incomplete requirement, and all we have are the past experience and data, so it is a wise decision to adopt EBA method because of its advantages.Estimation by analogy (EBA) in software cost estimation is an application of the theory of case based reasoning, and it uses information from former similar projects to predict the cost of a new project. Some advantages has been offered by analogy-based estimation: Firstly, it is easy to understand the basis of estimation. Secondly, it has the potential to mitigate problems with calibration. And lastly, it is useful where the domain is difficult to model.During the process of applying EBA to early estimation, some actual works have been done to promote this method.First, based on the four basic steps of analogy, an open frame which is used for the whole process of estimation by analogy has been developed. From the problem definition to the estimation validation, all of the steps are fully detailed.Second, the EBA method has been applied to early estimating, specified the range, content, and accuracy demand of the early estimation. At the same time, a few ways have been presented to upgrade the integrity of estimation, such as combining the validating mode with the predicting mode to take the advantages of the learning theory, and engaging the accuracy distribution function to analyze the impaction of the number of analogies and the threshold of similarity, et al. All of these methods were validated to be effective.Third, during the process of estimation practice, the software's size, effort, and duration have been succeedly measured in the early stage of software life circle, and these jobs proved that the improving methods are reasonable and doable. An optimize method on similarity calculation for large dataset has also been brought forward. This method can not only greatly reduce the amounts of calculation, but also avoid the effect of the less important factors, have improved the confidence estimating that on also certain degree.And next, the ways of measuring the effectiveness of the estimation process and the credibility of the data set have been colligated, also the evaluation criteria of the estimation method. And a questionnaire for the purpose of data collection was also developed, which may be the most important work for software estimation and measuring.From this thesis, the conclusion of the analogy method is an owing method can be concluded, which can be effectively used in the early stage of software estimation. So does it in every other stage of software life cycle.