Technical Debt Forecasting from Source Code Using Temporal Convolutional Networks

Technical Debt describes a deficit in terms of functions, architecture, or integration, which must subsequently be filled to allow a homogeneous functioning of the product itself or its dependencies. It is predominantly caused by pursuing rapid development versus a correct development procedure. Technical Debt is therefore the result of a non-optimal software development process, which if not managed promptly can compromise the quality of the software. This study presents a technical debt trend forecasting approach based on the use of a temporal convolutional network and a broad set of product and process metrics, collected commit by commit. The model was tested on the entire evolutionary history of two open-source Java software systems available on Github: Commons-codec and Commons-net. The results are excellent and demonstrate the effectiveness of the model, which could be a pioneer in developing a TD reimbursement strategy recommendation tool that can predict when a software product might become too difficult to maintain.