Systematic literature review on forecasting and prediction of technical debt evolution

TL;DR

This systematic review analyzes existing methods for forecasting technical debt evolution in software engineering, highlighting the dominance of machine learning techniques like random forests and temporal convolutional networks, but also noting significant gaps in coverage of debt types.

Contribution

The paper provides a comprehensive overview of current forecasting approaches for technical debt, identifying key techniques and gaps in the existing research landscape.

Findings

Random forest and temporal convolutional networks outperform other methods.

Current approaches only address two of fifteen technical debt types.

Research on TD forecasting is still in early development stages.

Abstract

Context: Technical debt (TD) refers to the additional costs incurred due to compromises in software quality, providing short-term advantages during development but potentially compromising long-term quality. Accurate TD forecasting and prediction are vital for informed software maintenance and proactive management. However, this research area lacks comprehensive documentation on the available forecasting techniques. Objective: This study aims to explore existing knowledge in software engineering to gain insights into approaches proposed in research and industry for forecasting TD evolution. Methods: To achieve this objective, we conducted a Systematic Literature Review encompassing 646 distinct papers published until 2023. Following established methodology in software engineering, we identified and included 14 primary studies for analysis. Result: Our analysis unveiled various approaches for TD evolution forecasting. Notably, random forest and temporal convolutional networks demonstrated superior performance compared to other methods based on the result from the primary studies. However, these approaches only address two of the fifteen identified TD types, specifically Code debt and Architecture debt, while disregarding the remaining types. Conclusion: Our findings indicate that research on TD evolution forecasting is still in its early stages, leaving numerous challenges unaddressed. Therefore, we propose several research directions that require further investigation to bridge the existing gaps. Keywords: Systematic literature review, Technical debt, Technical debt prediction, Technical debt forecasting, Technical debt metrics