Research On Software Project Delay Prediction Based On Small Sample Learning

With the advent of the new era of digital China and digital government,the number and scale of software projects are getting larger and larger,and the characteristics of knowledge and complexity are becoming more and more significant.However,software development is a complex ongoing system engineering,any problem in any part may lead to the failure of the entire project,strengthening the process management of software development projects is particularly important.For this reason,software development units are actively introducing software project management into software development to ensure the successful completion of software projects through effective project management.At present,the high failure rate of software projects has been widely concerned by domestic and foreign,through information technology means of real-time tracking of the project and scientific and reasonable progress estimates,to achieve early identification of project extension risks,early detection,early warning,to provide a scientific basis for the project to deal with the postponement of the formulation of timely remedial and adjustment measures to ensure the sustainable development of the software industry and innovation.Through the preliminary research and analysis,this thesis found a series of key problems such as small sample size of the data set used,unbalanced data,and difficulty in determining and quantifying the factors affecting software project extension.Therefore,this thesis focuses on the research of small sample size,unbalanced data set and quantification of software project extension factors,and the research focuses on the research of software project extension prediction model and algorithm.At present,there are several potential challenges in the more common software project extension prediction models,such as small sample data sets are difficult to adapt to the model,unbalanced data sets face poor prediction performance of traditional models,and software project extension impact factors are difficult to comprehensively analyze and evaluate.Therefore,this thesis improves the software project delay prediction model by improving the algorithm.To address the problems of small sample size,data imbalance and large impact of software project extension,this thesis carries out algorithm innovation research and model optimization research.For the problem of small sample size,the small sample learning algorithm based on transfer learning is studied,and the transferable knowledge transfer learning model is innovatively applied to the field of software project delay prediction.The MAML（model-independent meta-learning）algorithm is improved for the data imbalance problem,and the improved algorithm has higher performance than the existing imbalance processing algorithm through comparison experiments,with the highest BACC（balance of accuracy）value of 0.988.Meanwhile,the performance of preventing overfitting is significantly improved compared with other imbalance processing algorithms;the original data of software project delay prediction is processed by the above two improved algorithms to form a data set with sufficient samples and balance.On this basis,the software extension risk factors are determined and quantified,and the research constructs the software project extension prediction model Meta-IP,which realizes the quantitative prediction of software project extension,so as to judge the software project extension risk level.The experimental validation is based on software project data from a provincial department in Shandong province,and the experimental results show that the Meta-IP model studied in this thesis has the optimal prediction performance under the calibration of random forest classifier,and the AUC reaches 0.89 in 3^rd-year,which is 0.42 higher compared with the unprocessed data set.at the same time,it also achieves the effective validation.The research in this thesis provides an algorithmic model and practical validation for effectively carrying out software project extension prediction.