Compliance-Aware and Explainable GA-Optimized Neural Network for Cost Estimation in Safety-Critical Medical Software

Accurate cost estimation plays a crucial role in the development of medical software, given the safety-critical nature of these projects and their stringent compliance requirements. Traditional cost estimation methods, which often rely on algorithmic models, struggle to capture the complex nonlinear interactions among various cost drivers. On the other hand, while machine learning models have shown promise in handling such complexities, their black-box nature creates challenges in interpretability. This lack of transparency limits their acceptance and trustworthiness, especially within highly regulated environments where auditability and regulatory compliance are paramount. To overcome these challenges, we introduce GA-BP-XAI, a novel framework that combines a backpropagation neural network with genetic algorithm–based hyperparameter optimization, enhancing model performance while ensuring explainability. GA-BP-XAI integrates state-of-the-art interpretability techniques, specifically SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations), to provide transparent and auditable predictions. This approach facilitates a deeper understanding of the underlying cost drivers and supports compliance with regulatory requirements by enabling stakeholders to scrutinize and trust the model's outputs. The effectiveness of GA-BP-XAI was validated using a comprehensive dataset consisting of 1,200 anonymized medical software projects. Compared to a standard backpropagation neural network, GA-BP-XAI achieved an 11.6% reduction in mean absolute error (MAE) and improved the coefficient of determination (R²) from 0.902 to 0.927. Furthermore, it outperformed strong baseline models, including Linear Regression, Random Forest, and XGBoost, demonstrating superior predictive accuracy in cost estimation for medical software development. Beyond accuracy, explainability results from the GA-BP-XAI framework revealed key domain-relevant cost drivers such as FunctionPoints, ComplianceLevel, and IntegrationComplexity. These factors align well with expert knowledge in the field, underscoring the model’s ability to produce insights that are both meaningful and actionable. Such transparency is invaluable in compliance-driven contexts where stakeholders must understand not only what the predictions are but also why they arise, enabling more informed decision-making. In summary, GA-BP-XAI successfully bridges the gap between predictive performance and interpretability in cost estimation for safety-critical medical software. By delivering regulatory-aligned transparency alongside state-of-the-art accuracy, this framework offers a trustworthy tool for project managers, engineers, and regulatory bodies operating in high-stakes environments. This dual focus on precision and explainability ensures that medical software projects can be planned and executed with greater confidence, ultimately supporting safer and more cost-effective healthcare technology development.