Intelligent Inference in Combat Simulation Systems Based on Key Feature Extraction and Uncertainty Interval Estimation

Combat simulation has become crucial in military assessment in recent years due to the rapid development of information technology and artificial intelligence. However, the increasingly large volume of data and the credibility of information have heightened the difficulty of processing and analysis, subsequently affecting the quality of military decision-making. Extracting key features can clarify the core factors influencing the indicators, simplify model complexity, improve prediction accuracy, and save computation time. Quantifying uncertainty helps enhance decision quality and increases the system’s adaptability in uncertain environments. Accordingly, we propose a novel method for key feature selection and interval prediction to address specific regression tasks in combat simulation systems. First, our approach comprehensively considers the importance of features to the target variable, the interaction between features, and redundancy by integrating various feature selection methods, thereby precisely extracting key features. Second, we modify the output structure of traditional neural networks and design a new hybrid loss function to train the model. Furthermore, deep ensemble methods are utilized to enhance diversity and robustness, thus enabling uncertainty evaluation and interval prediction. The experimental results indicate that, after feature selection, the estimation achieved a mean squared error (mse) of only 0.151 and a prediction interval coverage probability (PICP) of 86.99%, providing crucial support for military decision-making.