Machine learning for software engineering: examples of two applications with ticket issues and code metrics

Abstract

In software engineering, tooling is nowadays mandatory for efficient project management and may relates to any project phase: design, coding or maintenance. These various tools consist of measures, called metrics, automated code generation systems and decision support systems, predictive or not. The applications of the latter are numerous: bad smell detection, ticket classification, etc. In this context of tooling, the use of machine learning methods in software engineering is constantly increasing. This is mainly due to the availability of data through online repositories but also due to the increase in computing power of machines.

In this talk I will present parts of my PhD concerning the use of machine learning techniques in the field of in software engineering. I will focus my presentation on two kinds of data where machine learning has been applied:

I) Issues: binary classification solution for discriminating bug from non bug issues. This solution combines natural language processing (TF-IDF) and classification (multi-layer perceptron) techniques, selected after comparing commonly used solutions to classify issues. Moreover, hyper-parameters of the neural network are optimized using a genetic algorithm. The obtained results, as compared to existing works on a commonly used benchmark, show significant improvements on the F1 measure for all datasets.

II) Metrics: approach to classify automatically developers using software metrics (senior or not). It uses a supervised learning-based approach trained with a manually labeled dataset of 703 developers from 17 open-source projects from GitHub for which 23 metrics are automatically extracted. Experienced developers classification results show a high F1 measure. A companion explainability study analyzes which metrics are the most influential.