Research And Application Of Software Testability Measurement Based On Fusion Model And Transformer

In modern society,the demand for software products is increasing day by day,and computer software is gradually applied to all aspects of social production and life.The complexity of software is also increasing,which greatly increases the difficulty of software testing.In order to effectively improve the efficiency of testing,reduce the difficulty and cost of testing,it is necessary to systematically analyze and measure the testability of software.This topic studies the testability measurement model of object-oriented software based on fusion model and Transformer respectively,which can objectively evaluate the testability of software.The model is also applied to the testability evaluation of new software projects of intern companies to provide a basis for improving software quality.The main work and research innovation of this thesis include the following aspects:1.Research on software testability parameters and measurement setIn this thesis,the testability of software is analyzed systematically and the scale of test suite required to achieve full coverage is taken as the software testability parameter.By collecting the real data of the enterprise software "force actuarial simulation Development system v1.0",preprocessing and analyzing it,and building the software testability measurement set with class as the research data object.2.Research on construction and application of software testability measurement model based on fusion modelIn this thesis,a single model and an integrated learning model are fused with the Stacking algorithm to predict software testability.Real data collected in the Force Actuarial Simulation Development System v1.0 were used as sample data to train Multiple Linear Regression,Support Vector Machines,Random Forests and XGBoost,respectively,and the algorithms in the Stacking algorithm were integrated.The experimental data show that the software testability measurement model based on fusion model is superior to other four machine learning models in performance.Finally,using the fusion model obtained by the company,the newly developed software "military simulation platform" and "military actuarial simulation development system v2.0" are taken as examples to explore the engineering application and predictive performance of the model.3.Construction and application research of testability measurement model based on TransformerConsidering the limitations of fusion model in the testability prediction of newly developed software,this thesis conducts pin-to-pair correlation analysis of class feature attributes and builds a lightweight Transformer model with the help of self-attention mechanism to measure the testability of software.After model training and comparative analysis with fusion model,the overall performance of transformer model is improved by10.82%,and the new model has higher stability.The effectiveness of the Self-Attention layer was verified by the ablation experiment of the model.The new model was also applied to the above-mentioned "Military Simulation Platform" and "Force Actuarial Simulation Development System v2.0" for engineering application research,and the application results show that Transformer-based software testability metric model can better solve the problem of unstable prediction effect of fusion model and has stronger testability prediction ability.