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

Abstract Accurate cost estimation is vital for medical software projects due to their safety-critical nature and strict compliance requirements. Traditional algorithmic models fail to capture nonlinear cost-driver interactions, while black-box machine learning approaches lack interpretability, limiting adoption in regulated contexts. To address this, we propose GA-BP-XAI, an explainable backpropagation neural network framework with genetic algorithm–based hyperparameter optimization. The framework integrates SHAP-and LIME-based interpretability to deliver transparent, auditable predictions. Using a dataset of 1,200 anonymized medical software projects, GA-BP-XAI reduces MAE by 11.6% and improves R 2 from 0.902 to 0.927 compared with standard BP, outperforming strong baselines such as Linear Regression, Random Forest, and XGBoost. Explainability results highlight domain-relevant drivers including FunctionPoints, ComplianceLevel, and IntegrationComplexity, consistent with expert knowledge. These results demonstrate that GA-BP-XAI achieves both state-of-the-art predictive accuracy and regulatory-aligned transparency, supporting trustworthy decision-making in compliance-driven, high-stakes environments.