A trustworthiness assessment for software based on multi-source evidence fusion

Software trustworthiness assessment plays a critical role in ensuring the safety, reliability, and accountability of software systems, especially in safety-critical and mission-critical applications, which is inherently complex, as the multi-source evidence is used across various stages of the software life-cycle. To address the resulting heterogeneity of evidence sources, this paper proposes a multi-stage and multi-source trustworthy evidence fusion method. First, a conflict measurement method between direct and indirect evidence is developed based on the concept of falsity. Second, a conflict resolution strategy for homogeneous evidence is proposed by integrating credibility and uncertainty metrics. Finally, a weighted combination method grounded in game theory is proposed to integrate falsity, credibility, and uncertainty, followed by evidence fusion using the Dempster rule. Experimental results in real software systems assessment demonstrate that the proposed model and algorithm not only satisfy the fundamental principles of trustworthiness assessment, namely monotonicity, acceleration, sensitivity, and substitutability, but also achieve higher accuracy and robustness compared with state-of-the-art algorithms. • A software trustworthiness assessment model is proposed to address multi-source heterogeneity in evidence fusion by distinguishing direct and indirect evidence. • A conflict quantification method based on falsity, credibility, and uncertainty metrics is developed to quantify and correct conflicts in multi-source evidence. • A game-theoretic weighting strategy is designed to generate composite discounting factors for adjusting evidence values in the fusion process. • Assessment on real software systems shows the proposed model meets key trustworthiness principles and performs better than traditional evaluation models.