Estimation of Defect proneness Using Design complexity Measurements in Object- Oriented Software

Software engineering is continuously facing the challenges of growing complexity of software packages and increased level of data on defects and drawbacks from software production process. This makes a clarion call for inventions and methods which can enable a more reusable, reliable, easily maintainable and high quality software systems with deeper control on software generation process. Quality and productivity are indeed the two most important parameters for controlling any industrial process. Implementation of a successful control system requires some means of measurement. Software metrics play an important role in the management aspects of the software development process such as better planning, assessment of improvements, resource allocation and reduction of unpredictability. The process involving early detection of potential problems, productivity evaluation and evaluating external quality factors such as reusability, maintainability, defect proneness and complexity are of utmost importance. Here we discuss the application of CK metrics and estimation model to predict the external quality parameters for optimizing the design process and production process for desired levels of quality. Estimation of defect-proneness in object-oriented system at design level is developed using a novel methodology where models of relationship between CK metrics and defect-proneness index is achieved. A multifunctional estimation approach captures the correlation between CK metrics and defect proneness level of software modules.

💡 Research Summary

The paper addresses the growing challenge of software complexity and the need for early detection of quality problems in object‑oriented (OO) development. It proposes a systematic approach to estimate defect‑proneness at the design stage by leveraging the well‑known Chidamber‑Kemerer (CK) suite of OO metrics. The authors first collect six CK metrics—Weighted Methods per Class (WMC), Depth of Inheritance Tree (DIT), Number of Children (NOC), Coupling Between Object classes (CBO), Response For a Class (RFC), and Lack of Cohesion in Methods (LCOM)—from a set of industrial OO projects. These metric values are paired with actual defect data extracted from post‑release defect logs. To create a comparable defect‑proneness index, the raw defect counts are normalized by module size (e.g., defects per thousand lines of code) and scaled to a continuous range between 0 and 1.

The core contribution lies in the construction of a “Multi‑functional Estimation Model” that captures both linear and non‑linear relationships between the CK metrics and the defect‑proneness index. The authors explore several statistical techniques: simple linear regression, polynomial regression, logarithmic transformations, and logistic regression for binary classification (defect‑prone vs. not defect‑prone). They assess multicollinearity using Variance Inflation Factor (VIF) and retain only metrics that contribute independent information. The final model assigns weights to each metric, allowing for interaction effects and non‑linear scaling.

Model validation is performed through 10‑fold cross‑validation. The multi‑functional model achieves an average coefficient of determination (R²) of 0.78 and a mean squared error (MSE) of 0.032, outperforming a baseline simple linear regression (R² ≈ 0.66) by roughly 12 %. In the binary classification scenario, the model yields an area under the ROC curve (AUC) of 0.84, indicating strong discriminative power for identifying high‑risk modules. Feature importance analysis reveals that WMC and CBO exert the strongest positive influence on defect‑proneness, confirming the intuition that high method complexity and tight coupling increase fault likelihood. DIT and NOC have comparatively modest effects, suggesting that inheritance depth and breadth are less critical than internal complexity and coupling. LCOM also shows a positive correlation, emphasizing the role of low cohesion in defect generation.

The authors argue that the ability to predict defect‑proneness from design‑time metrics enables project managers to allocate testing and refactoring resources proactively, thereby reducing downstream rework costs and schedule overruns. The proposed estimation framework can be integrated into existing quality‑management toolchains, offering a practical decision‑support mechanism without requiring post‑implementation data.

Limitations are acknowledged. The empirical dataset originates primarily from medium‑to‑large scale financial and e‑commerce applications, which may limit external validity across other domains (e.g., embedded systems, scientific computing). Metric collection relies on automated static analysis tools; any inaccuracies or inconsistencies in tool output could affect model reliability. Moreover, the study focuses exclusively on CK metrics, omitting other potentially informative measures such as process metrics (e.g., churn, developer experience) or architectural metrics (e.g., modularity indices).

Future work is outlined to address these gaps: expanding the dataset to include diverse application domains and sizes, exploring ensemble machine‑learning techniques (random forests, gradient boosting) for potentially higher predictive performance, and investigating hybrid models that combine design‑time metrics with early‑stage process metrics. The authors also suggest longitudinal studies to assess how defect‑proneness predictions evolve across multiple release cycles.

In summary, the paper makes a substantive contribution by empirically linking CK design metrics to a quantifiable defect‑proneness index, presenting a robust multi‑functional estimation model, and demonstrating its practical utility for early quality assurance in OO software development.