Intelligent Detection of Open Source Software Anomalies

点击翻译按钮获取中文摘要

Software that is not fully tested, can show unexpected results during execution that is referred as Software Anomaly (SA). Specific SA is known as Implied Scenario (IS): a new scenario is implied during execution time. A popular example of IS is a boiler system in which the control component sends commands to an actuator based on the previous collected data from sensors instead of the current data. The earlier fixing of IS reduces the costs. Hence, IS research rely on studying IS from Sequence Diagrams (SD). Most research require human expertise (e.g. to annotate SD), which is error-prone and time consuming. These approaches cannot help developers to avoid IS while developing a software. Also, in practice, SD is not used or documented by companies. On the other hand, there is much research that build SA prediction algorithms (using anomaly reports data with metrics such as lines of code added) to determine parts of software that are more probable to have SA. However, these works do not directly address IS. Objective. The short-term objective of this research program is to develop heuristic-based techniques and tools for IS prediction using the SA prediction metrics, which will help software developers to automatically analyze the code for IS and visualize the results while developing software. Open Source Software (OSS) is the main focus of this research due to the increasing number of companies that are adopting OSS solutions, while there is a need to support companies in managing and integrating the OSS. Methods. In the first stream of the program we will investigate the SA prediction metrics that can be used for IS and develop an IS prediction metrics database. We will conduct systematic literature review and explore public SA report datasets to develop an IS library. We will also study SA prediction research and use topic modelling to develop an IS Prediction Metrics Database for the IS library. Secondly, we will build IS prediction models based on heuristic approaches. We will explore different algorithms and metrics to increase IS prediction accuracy. Also, we will study automation of machine learning approaches to develop models that choose prediction metrics automatically. Finally, we will study visualization techniques and OSS design patterns to build open source tools for analyzing code for IS. We will evaluate our models with various datasets using our prototypes and compare results with the current IS prediction works. The usefulness of the tool will be surveyed online for developers worldwide. HQP and Impact. I expect practical and theoretical contributions in IS prediction research that will benefit Canadian software industry in producing high quality software (i.e. avoiding IS) and companies in integrating OSS solutions. The program will train Highly Qualified Personnel (HQP) with expertise in data science and software engineering. The HQP will publish in top-ranked peer-reviewed venues as the first author.