The Research And Implementation Of Case-Based Reasoning To Estimation By Analogy

Accurate prediction is one of the most important goals for the software engineering community. Most work has focused upon building algorithmic models, for example COCOMO. These must be calibrated to local environments. Another approach is estimation by analogy that can confirm the value of estimate of the target case by fact experience in past, is more easily understandable than the algorithm model.We study and analyse the method of the CBR(Case-Based Reasoning) and machine learning and estimation by analogy. We analyse the method of estimation by analogy by the principle and technology of CBR, and bring forward an estimat method that contains two stages and four steps. Our method divide estimation by analogy into project estimate stage and retain case stage, and we achieve estimating target case by analogical case selection, resolving reuse and resolving revise. Testing the accuracy of estimate in the data set and adjusting value of estimate for the next estimate at retain case stage.We attempt to analysis the correlation between attributes of the data set by statistical regression in the method of estimatoin by estimation, and find out most relevant attributes of the target attribute to reduce the MMRE. In the solution process, we using Euclidean distance squared reciprocal weighting method to diminish the relation between the number of analogical cases and the value of performance indication to improve the accuracy of the estimate.On this basis, we designed and imlemented software tool to estimate by analogy to select the most relevant attributes, select analogical case and solve the value of estimate.the tool can not only do authentication method such as Jacknife, n-fold, but also calculate the value of performance indication, such as MMRE, Pred(x).We analyse the desharnais and ISBSG by the tool, and educe that the most relevant attributes selection and multi-project synthesis method can greatly improve the accuracy of estimates, and distance squared reciprocal weighting method can diminish the relation between the number of analogical cases and MMRE effectively and eliminate the change of selecting the most relevant attributes the impact of the different values of N. The tool can observe the relation between the number of analogical cases and the value of performance indication to determine the most relevant attributes more directly than ANGEL software tool.We have show that estimation by analogy is an objective, accurate and applicable method, and make up the insufficiency of other estimation methods.