The QuaDeeP (Enhancing the quality of software systems using deep learning models for defects prediction and detection) project focuses on developing deep learning techniques for software defect prediction, a problem of major relevance within software engineering, and particularly within search-based software engineering. The major goal is to improve the quality of software systems by early and accurate identification of defective software modules, using models and techniques from deep learning. The project topic is of major international interest, being of great relevance during the development, testing and maintenance of software systems. Accurate prediction of software defects in new software versions would significantly improve the performance of the software development process in terms of cost, time and software quality.
Project Outcome
The principal expected project outcome will be the QuaDeeP software solution which will integrate novel deep learning methods for software defects identification. For increasing the specificity of the models, the targeted learning methods will be specifically tailored for particular types of defects. QuaDeeP will be useful for assisting software developers in accurately predicting software defects, thus contributing to improve software quality and to ease software maintenance and evolution.
Objectives
O1. Development and scientific validation of novel deep learning based methods for the feature engineering step for software defect prediction
We will employ existing taxonomies of defect types such as ODC, CWE or the CVE to identify relevant features which are specific to particular classes of defects. Machine learning models such as autoencoders, CNNs and LSTMs will be targeted to automatically learn semantic and syntactic features from source code representations. New cohesion and coupling based software metrics for defect prediction will be expressed based on existing software metrics and semantic source code representations.
O2. Development and scientific validation of novel machine learning based models and techniques for software defect prediction
Machine learning models will be tailored for particular defect types, increasing model specificity in learning to predict a particular classes of defects. We envisage using one-class classification and one-shot learning methods for handling the main issue of data imbalance.
O3. Development and validation of the QuaDeeP software modules
Provided in the form of software modules, QuaDeeP will provide a solution to assist developers, testers and software managers in activities related with software maintenance and evolution, by providing information that allows stakeholders to pinpoint possible defects in software.
O4. Contribute to the development of scientific knowledge by disseminating the obtained scientific results through scientific publications and the project website
Project financed under PN-III-P4-ID-PCE-2020-0800, contract nr. PCE 92/2021