Discovering software developer's coding expertise through deep learning

• Published in: IET software Vol. 14; no. 3; pp. 213 - 220
• Main Authors: Javeed, Farooq, Siddique, Ansar, Munir, Akhtar, Shehzad, Basit, Lali, Muhammad I.U
• Format: Journal Article
• Language: English
• Published: The Institution of Engineering and Technology 01.06.2020
• Subjects: appropriate developer; expert developers; learning (artificial intelligence); novice; project management; public domain software; software developer coding expertise; software developers; software development management; software development organisations; software engineering; Special Issue: Knowledge Discovery for Software Development (KDSD); successful software development; project management; expert developers; successful software development; software developer coding expertise; public domain software; appropriate developer; software development management; software development organisations; software developers; software engineering; learning (artificial intelligence); novice
• ISSN: 1751-8806; 1751-8814; 1751-8814
• DOI: 10.1049/iet-sen.2019.0290

The field of software development is growing rapidly and prevailing in every walk of life. The role of software developers in such a challenging and complex activity is very much important. The allocation of right software developers (i.e. who possesses appropriate coding skills) to projects is one of the crucial factors for successful software development. The problem is that it is very difficult for a client, project manager, as well as for software development organisations to find out an appropriate developer and assign him/her to a particular project. To achieve this, there is a need for such a sound mechanism that could detect the level of software developer coding expertise. This study has formulated criteria for novice and expert developers and carried out such criteria to discover the level of coding expertise of software developers using three different models of deep learning. These models include long short-term memory (LSTM), convolution 1D and hybrid (a combination of LSTM and convolution 1D). The deep learning models have analysed software developers’ previously written source code collected from the GitHub repository. An experiment was conducted to evaluate the performance of models. The results showed that the LSTM model performed better in comparison to other models by achieving 96.25% accuracy.