Research On Software Module Defect Prediction Method In Fire Maintenance System

With the rapid development of the Internet,software has now played an increasingly important role in all walks of life,and people have higher and higher requirements for software quality.Therefore,how to quickly and accurately find potential defects in software has caused more and more attention.At the same time,the quality of the fire maintenance system independently developed by the laboratory has become increasingly prominent,and its testing became more and more difficult.Therefore,this paper is oriented to the fire maintenance system,and designs a software module defect prediction method based on RFP-TCS-XGBoost,which is used to determine the modules that may have defects in the fire maintenance system.The work of this paper comes from the fire maintenance system.Because the system's code quality problems and existing software defect prediction research are still more based on open data sets,which cannot be well combined with the characteristics of the actual project itself,the software module defect prediction method in the fire maintenance system was studied.The main work includes the following four aspects:(1)Collect the data set of fire maintenance system,design the measurement elements suitable for fire maintenance system,and propose NPM,ADC and MI measurement elements.By analyzing and collecting the data of the fire maintenance system,according to the characteristics of the system,a set of metric elements combining software static factors and dynamic development process factors is proposed,and three special metrics for the fire maintenance system were designed.In the end,formula design and data collection are performed for the designed metric.(2)According to the characteristics of the fire maintenance system data set,complete the data preprocessing work,propose a two-stage feature selection method of RFP,and screen out a feature set composed of 8 metric elements.In the data preprocessing part,by analyzing the characteristics of the metric elements of the data set,the feature coding of categorical variables is completed;for the problem of too large dimensional differences between each metric element,the data standardization work is completed;In addition,a two-stage feature selection method RFP based on ReliefF weight analysis and Pearson correlation analysis is designed to comprehensively consider the correlation between each metric and the target features and eliminate the redundancy of the metric.The set of metric elements are highly correlated but not redundant with the target feature.(3)Investigate the software module defect prediction model TCS-XGBoost for fire maintenance system.By introducing cost-sensitive ideas,combined with the data set and business process characteristics of the fire maintenance system,a three-tier cost-sensitive software module defect prediction model called TCS-XGBoost was proposed to predict the defect tendency of software modules.(4)Implement and test each module of RFP-TCS-XGBoost to verify the effectiveness of the software module defect prediction method proposed in this paper.In this paper,relevant evaluation indexes including accuracy rate,precision rate,recall rate,FI value,ROC curve and AUC area are selected,and experiments in three aspects are carried out.Firstly,a comparison test of whether the design includes three unique metric elements shows that the NPM,ADC and MI metric elements proposed in this paper can significantly improve the model's defect prediction effect.Then the feature selection experiment is designed to show the effectiveness of the two-stage feature selection method RFP designed in this paper.Finally,through a comparative analysis with the decision tree DT and XGBoost algorithms,it is shown that the software module defect prediction model called TCS-XGBoost proposed in this paper has better performance on the fire maintenance system.