Empirical validation of human factors in predicting issue lead time in open source projects

  • Authors:

  • Nguyen Duc Anh;Daniela S. Cruzes;Reidar Conradi;Claudia Ayala

  • Affiliations:

  • IDI-NTNU Trondheim, Norway;IDI-NTNU Trondheim, Norway;IDI-NTNU Trondheim, Norway;Technical University of Catalunya, Barcelona, Spain

  • Venue:
  • Proceedings of the 7th International Conference on Predictive Models in Software Engineering
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

[Context] Software developers often spend a significant portion of their resources resolving submitted evolution issue reports. Classification or prediction of issue lead time is useful for prioritizing evolution issues and supporting human resources allocation in software maintenance. However, the predictability of issue lead time is still a research gap that calls for more empirical investigation. [Aim] In this paper, we empirically assess different types of issue lead time prediction models using human factor measures collected from issue tracking systems. [Method] We conduct an empirical investigation of three active open source projects. A machine learning based classification and statistical univariate and multivariate analyses are performed. [Results] The accuracy of classification models in ten-fold cross-validation varies from 75.56% to 91%. The R2 value of linear multivariate regression models ranges from 0.29 to 0.60. Correlation analysis confirms the effectiveness of collaboration measures, such as the number of stakeholders and number of comments, in prediction models. The measures of assignee past performance are also an effective indicator of issue lead time. [Conclusions] The results indicate that the number of stakeholders and average past issue lead time are important variables in constructing prediction models of issue lead time. However, more variables should be explored to achieve better prediction performance.