Comparativo de Arima e SVM na previsão de risco em teste de software

Abstract

Software testing has been playing an increasingly important role in the life cycle of computer systems. Quality assurance has become a requirement and with the obstacles of strict deadlines and scarce human and financial resources, methods that offer a greater test coverage and identification of problems in the shortest time available have been sought. The present work conducted a practical study that combines Risk Based Testing, an approach that executes only carefully selected test cases based on the most critical modules and functionalities of a system, applied with complete tests, with risk prediction based on analysis of time series by applying Support Vector Machine (SVM) and Autoregressive Integrated Moving Average (ARIMA) models to optimize test script selection and prioritization. The results showed an improvement in the perception of quality by the team and by the end users, with a reduction in the number of raised problem tickets. Additionally, the models did not obtain high values for the metrics of accuracy, precision, recall, RSME and MAE. However, ARIMA achieved better results compared to SVM.