Research And Implementation Of Software Cost Estimation Method Of COCOMO ? Improved Model

With the rapid development of the software industry,software cost management,as an important part of software engineering,has received more and more attention from relevant practitioners.The COCOMO ? model is the most extensively researched software cost estimation model.Its success is due to the strict isolation of the factors affecting software cost,expressed as 5 scale factors and 17 cost driving factors,each of which is divided into four to six levels.How to use these factors accurately and scientifically is a subject worthy of longterm discussion and development.With the rise of machine learning and the precipitation of historical data,it is possible to study the grading process of these factors in depth.The mainly study of this paper is the software complexity and reliability that have a significant impact in the COCOMO ? model.The main work of this paper is as follows:(1)In the COCOMO ? model,the complexity classification is subjective,difficult to operate,and its evaluation level is too high.This paper proposes a software complexity classification method based on module.First,according to the module breakdown structure,the software is divided into multiple modules from top to bottom,and the characteristics of each module are divided into qualitative characteristics and quantitative characteristics according to their own nature.The gray clustering and quantitative analysis of complexity classification methods are respectively proposed,and The complexity grading results of the two characteristics are synthesized by the multi-weighted evaluation method to obtain the complexity level of each module.Finally,according to the module breakdown structure,the software complexity level is comprehensively evaluated from the bottom up,and applied to the improved COCOMO ? model.This paper takes MI system,an operator's software with intelligent recommendation function as an example,gives the software complexity grading process,and verifies the feasibility of the above method.(2)This paper aims to improve the software reliability grading process in the COCOMO ? model based on defect prediction.This method can integrate the reliability of the software itself and the required reliability level of the software.First,a defect propensity prediction model based on CNN-SVM is proposed for multi-dimensional defect features of modules,and a defect number prediction model based on SVR is proposed for fine-grained defect features.These two data sets are verified on PC5 and CK data sets.The performance of each model,and then the above prediction results are counted to obtain the module defect rate or the number of defects per thousand code lines,and then the reliability of the software itself is rated.The software required reliability level is graded by analogy to the failure loss of the software type.Finally,taking the MI system as an example,the improved software reliability grading process is given,and the feasibility of the above method is verified.(3)Based on the above research,a software cost estimation system based on the improved COCOMO ? model is realized.The overall design of the system and the specific functions realized by each module are introduced in detail,and displayed through the interface.Finally,a software cost estimation case of MI system is given to verify the method in this paper.