Third-party library selection in IT projects under imperfect data using Dempster–Shafer theory

Abstract

This study focuses on a method for selecting third-party libraries for IT projects, which involves systematizing evaluation criteria and applying Dempster–Shafer theory of evidence to model imperfect data. Imperfect data refer to incomplete, contradictory, unreliable, uncertain, imprecise, or ambiguous information. The goal is to enhance the validity and reliability of library selection and replacement decisions in the ever-changing landscape of modern projects while also minimizing associated risks. The tasks to be solved include: providing a systematic classification of evaluation criteria that comprehensively characterize third-party libraries; developing a selection method of third-party libraries that incorporates evaluations as imperfect data; integrating and combining evaluations from multiple heterogeneous sources; and demonstrating the application of the proposed method through an illustrative example that establishes confidence intervals for alternative libraries. The methods used are based on the Dempster–Shafer theory of evidence for modeling imperfect data and the Dubois–Prade disjunctive consensus rule for combining evaluations from independent sources. The results show that the proposed approach transforms subjective and imperfect evaluations into evidence, combines them according to the selected rule of evidence theory, and derives confidence intervals that express both guaranteed and possible degrees of support for each library alternative. This study confirms the effectiveness of applying Dempster–Shafer theory of evidence in multi-criteria decision-making contexts that resemble real-world project environments. Conclusions. The scientific novelty of this study lies in proposing, for the first time, a method for selecting third-party libraries based on the Dempster–Shafer theory of evidence, distinguished by a systematic taxonomy of evaluation criteria, including risk factors, and by combining evidence in support of candidate tools using the Dubois–Prade disjunctive consensus rule. The developed method extends the analytical capabilities of project decision-support systems by enabling comprehensive evaluation and risk-informed selection of third-party libraries in complex, dynamically evolving technological environments.