Abstract
In the last few years, the increasing trend of vessel density, different types of vessels, and the increased need for real-time data have made maritime traffic management significantly more difficult. This study presents a situation-aware framework based on stacked ensemble learning and cloud-edge hybridization, which is aimed at enhancing the maritime traffic monitoring and control systems. This approach combines stacked ensemble learning with a meta-model for vessel type classification and employs the concept of cloud-edge architecture to strike a balance between computational efficiency and delay minimization. While the edge layer takes care of real-time inference and situational analysis on the go, the cloud layer takes care of model training and amalgamation of data from various sources. Our evaluation made use of a comprehensive maritime vessel dataset and compared the performance with the state-of-the-art deep learning models (VGG16, VGG19, DenseNet121, and ResNet50). Our experiments show that the stacked ensemble learning with a meta-model significantly outperforms the traditional ones, achieving an overall accuracy of 0.98, macro average precision of 0.97, macro average recall of 0.98, and an F1-score of 0.98. Both ROC and PR curves also demonstrate excellent AUC values, which tend to 1.00 for almost all categories of vessels, which is a strong performance in distinguishing vessels from each other. Test predictions are outstandingly accurate, with confidence in vessel classification exceeding 99% in most cases. From these results, the proposed method shows robustness, scalability, and effectiveness for real-time maritime surveillance, naval defense systems, and autonomous vessel traffic control in industrial IoT environments.