Machine learning techniques for software testing effort prediction

Software testing (ST) has been considered as one of the most important and critical activities of the software development life cycle (SDLC) since it influences directly on quality. When a software project is planned, it is common practice to predict the corresponding ST effort (STEP) as a percentage of predicted SDLC effort. However, the effort range for ST has been reported between 10 and 60% of the predicted SDLC effort. This wide range on STEP causes uncertainty in software managers due to STEP is used for allocating resources to teams exclusively for testing activities, and for budgeting and bidding the projects. In spite of this concern, hundreds of studies have been published since 1981 about SDLC effort prediction models, and only thirty-one STEP studies published in the last two decades were identified (just two of them based their conclusions on statistical significance). The contribution of the present study is to investigate the application for STEP of five machine learning (ML) models reported as the most accurate ones when applied to SDLC effort prediction. The models were trained and tested with data sets of projects selected from an international public repository of software projects. The selection for projects was based on their data quality rating, type of development, development platform, programming language generation, sizing method, and resource level of projects. Results based on statistical significance allow suggesting the application of specific ML models to software projects by type of development, and developed on a determined platform and programming language generation.