Hybrid Evolutionary–Deep Learning Framework for High-Reliability Prediction in Safety-Critical Systems

Abstract

Safety-critical systems require highly reliable and intelligent prediction mechanisms to ensure operational safety, fault tolerance, and decision accuracy under dynamic environments. Conventional machine learning approaches often suffer from limited feature representation and insufficient optimization capability, leading to degraded predictive performance. To address these challenges, this paper proposes a Hybrid Evolutionary–Deep Learning Framework that combines evolutionary optimization with deep learning techniques for high-reliability prediction in safety-critical systems. Initially, data preprocessing and normalization are performed to remove inconsistencies and enhance feature quality. An evolutionary optimization algorithm is employed to identify the most informative features and optimize the hyperparameters of the deep neural network. The optimized deep learning model subsequently performs classification and prediction tasks with improved robustness and generalization capability. Extensive experiments are conducted using benchmark datasets relevant to safety-critical applications. The proposed framework achieves an accuracy of 98.76%, precision of 98.31%, recall of 98.54%, F1-score of 98.42%, and an area under the ROC curve (AUC) of 99.12%, outperforming conventional machine learning and standalone deep learning approaches. Furthermore, the framework demonstrates superior reliability and reduced prediction error, making it suitable for deployment in domains such as autonomous systems, industrial monitoring, healthcare diagnostics, and aerospace applications. The obtained results confirm that integrating evolutionary optimization with deep learning significantly enhances predictive performance and reliability in safety-critical environments.