A Study on Improvement of Training Data Quality Management for AI-based Intelligent Naval Combat Management System

Purpose: The purpose of this study was to suggest the metrics that manage the quality of train datasets for artificial intelligence in order to support quality management activities of intelligent naval combat management system. Furthermore, the study demonstrates how enhancing datasets can lead to measurable improvements in artificial intelligence performance based on the proposed metrics. It involves observing how quality control on the evaluated dataset relates to the performance of artificial intelligence.Methods: Practical datasets mimicking data structure of naval combat management system were constructed to reflect operational contexts of intelligent naval combat systems. The proposed 17 metrics for managing dataset quality were applied to these datasets for quantitative evaluation, called operation definition.Results: The metrics quantified key quality attributes of the training datasets for intelligent combat system. By using different quality datasets, we conducted metric-based and AI model-based experiments and found that the metrics are suitable for naval combat system applications. According to the two types of experiments, it was found that the model trained using a high-quality dataset demonstrated superior performance compared to the model trained on a low-quality one. Overall, these results support the possibility of applying our framework to real-world defense scenarios.Conclusion: Through the quality management activities considering the functions of AI-centric weapon system, it becomes possible to ensure data-level quality control that supports the stable operation of key functions within AI-based intelligent naval combat systems.