Research On Software Project Quality Prediction Based On Neural Network

Nowadays,all organizations are inseparable from the application of software,and the importance of software project quality is increasing day by day.It is usually difficult to make a more accurate prediction of software quality before project research and development.The lack of accurate quality prediction will bring great trouble to project managers and may lead to project failure.This paper uses neural network related technology to predict the quality of software project,in order to make the quality prediction of software R&D project more accurate through the training of neural network.Neural network and other machine learning methods are introduced into software projects to improve project managers’ control of project risks.The main idea is to build a regression model using neural network to predict the collected project related data.Collect the data related to software project R&D as samples,and fully realize the accurate prediction of software project quality through the six steps of neural network:data set,model,loss function,optimizer,training and prediction.Therefore,there are two possible innovations:one is the application of characteristic data,and the other is the accuracy of prediction results.Through the detailed explanation of the basic knowledge of neural network,including the form of data set,the introduction of basic components of neural network model such as neuron and activation function,the introduction of model measurement standard loss function,the detailed explanation of training steps of machine learning and the use of prediction.Combined with the current research on the quality prediction of relevant software projects,this paper makes a detailed comparison with the methods used in this paper.It is found that the problems existing in the current quality prediction methods of relevant software projects mainly include the lack of objectivity of the prediction model,the insufficient data mining of project stakeholders,and the accuracy of model prediction.The main reasons for these problems include the lack of consideration of the rationality of characteristic data Lack of pertinence to a certain research and development field,lack of theoretical explanation of neural network,lack of training and verification methods of neural network model.Finally,according to these problems and the causes of these problems,this paper puts forward improved countermeasures and methods,that is,collect the sample data sets of 4500 software R&D projects,preprocess the data by using the methods of unique heat coding,feature scaling and verification set,and then design a three-layer neural network model,using the mean square error loss function and random gradient descent method,Combined with the early stop method to prevent over fitting,an accurate conclusion for software project quality prediction is finally drawn.