Design-level software fault prediction using SOLID-based LSP and ISP violation metrics

Software fault prediction is essential for improving software reliability and reducing maintenance costs. Traditional metrics, including object-oriented, process, and complexity metrics, effectively capture structural and historical properties but often fail to identify design-level violations that contribute to software faults. To address this limitation, this paper proposes a hybrid fault prediction framework that integrates existing metrics with SOLID principle-based design metrics. The framework introduces two design violation metrics: the Liskov Substitution Principle Violation metric (LSPV) and the Unified Interface Segregation Principle metric (UISP), which quantify behavioral inconsistencies, interface misuse, and unnecessary coupling in object-oriented systems. Principal Component Analysis (PCA) is employed to reduce redundancy and multicollinearity by generating compact and information-rich feature representations. An integrated metric, the Hybrid Design Violation Metric (HDVM), probabilistically combines LSPV and UISP using a sigmoid-based formulation to estimate software fault-proneness. Existing benchmark datasets do not include SOLID principle-based design metrics. Therefore, a new dataset was constructed from three open-source Java projects (Apache Ant, Apache Lucene, and Apache Camel). To evaluate the fault prediction capability of HDVM, a controlled mutation-based dataset was further constructed by injecting representative LSP and ISP design violations while preserving the functionality and compilability. Experimental results and ablation analysis demonstrate that HDVM achieves superior predictive capability and better generalization than models relying solely on conventional metrics. Furthermore, bug-type analysis reveals that HDVM is particularly effective in identifying faults associated with inheritance misuse, polymorphism violations, interface inconsistencies, and architectural coupling degradation, highlighting the importance of design-level metrics for fault prediction.