Machine Learning for Technical Debt Identification

Tsoukalas, Dimitrios; Mittas, Nikolaos; Chatzigeorgiou, Alexandros; Kehagias, Dionysios; Ampatzoglou, Apostolos; Amanatidis, Theodoros; Angelis, Lefteris

doi:10.1109/tse.2021.3129355

Cited by 14 publications

(11 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…In the near future, we plan to include in the platform additional refactoring identification approaches. Furthermore, regarding monitoring, we have already incorporated forecasting capabilities 29 ; and for TD identification we have introduced a machine learning approach 30 that is able to identify the artifacts with high‐levels of TD, using the intersection of three well‐known tools (Sonar, CAST, and Squore) 10 …”

Section: Discussionmentioning

confidence: 99%

SDK4ED: A platform for technical debt management

et al. 2022

Self Cite

View full text Add to dashboard Cite

Technical debt management is of paramount importance for the software industry, since maintenance is the costlier activity in the software development lifecycle. In this article, we present the SDK4ED platform that enables efficient technical debt management (i.e., measurement, evolution analysis, prevention, etc.) at the code level, and evaluate its capabilities in an industrial setting. The SDK4ED platform is the outcome of a 3-year project, including several software industries. Since, the research rigor of the approaches that reside in SDK4ED have already been validated, in this work we focus: (a) on the presentation of the platform per se; (b) the evaluation of its industrial relevance; (c) the usability of the platform; as well as (d) the financial implications of its usage.

show abstract

Section: Discussionmentioning

confidence: 99%

SDK4ED: A platform for technical debt management

et al. 2022

Self Cite

View full text Add to dashboard Cite

show abstract

“…Most these models have been extensively compared and evaluated in the literature for their ability to predict important software attributes. 5,33,[45][46][47][48] In order to tune selected models in the best possible way, we used the grid-search method. 49 Grid search is commonly used to find the optimal hyperparameters of a model that result in the most accurate predictions, by performing an exhaustive search over specified parameter values.…”

Section: Model Constructionmentioning

confidence: 99%

“…Subsequently, a set of causal and ML models, namely, multiple linear regression (MLR), ridge and lasso regression, SVR with both linear and Gaussian kernel, and random forest regression were selected for a class‐level evaluation. Most of these models have been extensively compared and evaluated in the literature for their ability to predict important software attributes 5,33,45–48 . In order to tune selected models in the best possible way, we used the grid‐search method 49 .…”

Section: Empirical Studymentioning

confidence: 99%

A practical approach for technical debt prioritization based on class‐level forecasting

Tsoukalas

Siavvas

Kehagias

et al. 2023

J Software Evolu Process

Self Cite

View full text Add to dashboard Cite

Monitoring technical debt (TD) is considered highly important for software companies, as it provides valuable information on the effort required to repay TD and in turn maintain the system. When it comes to TD repayment, however, developers are often overwhelmed with a large volume of TD liabilities that they need to fix, rendering the procedure effort demanding. Hence, prioritizing TD liabilities is of utmost importance for effective TD repayment. Existing approaches rely on the current TD state of the system; however, prioritization would be more efficient by also considering its future evolution. To this end, the present work proposes a practical approach for prioritization of TD liabilities by incorporating information retrieved from TD forecasting techniques, emphasizing on the class‐level granularity to provide highly actionable results. Specifically, the proposed approach considers the change proneness and forecasted TD evolution of software artifacts and combines it with proper visualization techniques, to enable the early identification of classes that are more likely to become unmaintainable. To demonstrate and evaluate the approach, an empirical study is conducted on six real‐world applications. The proposed approach is expected to facilitate developers better plan refactoring activities, in order to manage TD promptly and avoid unforeseen situations long term.

show abstract

“…As machine learning had yet to be applied to the technical debt prioritization problem, as (Tsoukalas, Mittas, et al, 2021) and (Mauricio Aniche et al, 2020), we decided to apply simple methods to better understand the results. The nine selected methods are Dummy Classifier, Naive Bayes, K-Nearest Neighbors, Logistic Regression, Ridge Classifier, Support Vector Machine, Decision Tree, Random Forest, and XGBoost.…”

Section: Research Questionsmentioning

confidence: 99%

“…Tsoukalas et al (Tsoukalas, Mittas, et al, 2021) used well-known machine learning methods to classify software modules in high-TD or not. Our work uses almost the same machine learning methods, but we applied them to classify whether and when a technical debt item should be paid off.…”

Section: Related Work For Technical Debt Prioritization Using Machine...mentioning

confidence: 99%

Technical debt prioritization: methods, techniques, and a large exploratory study

Pina

View full text Add to dashboard Cite

show abstract

Machine Learning for Technical Debt Identification

Cited by 14 publications

References 38 publications

SDK4ED: A platform for technical debt management

SDK4ED: A platform for technical debt management

A practical approach for technical debt prioritization based on class‐level forecasting

Technical debt prioritization: methods, techniques, and a large exploratory study

Contact Info

Product

Resources

About