Identification of air targets using a hybrid clustering algorithm

The study addresses the task to improve the accuracy of clustering air raid danger levels by constructing a hybrid clustering algorithm. A target air clustering algorithm has been developed, which involves using a modified distance metric and integrates air danger level assessments directly into the algorithm. The reported features demonstrate superiority over existing algorithms based on the Silhouette and Davies-Bouldin indices. The proposed model yields a Silhouette index of 0.72306 compared to 0.3481 for the existing model, and a Davies-Bouldin index of 0.3389 compared to 1.209. Models such as Random Forest Classifier and Gradient Boosting Classifier, evaluated using the clusterizer, exhibit higher accuracy, specifically 0.87 and 0.87, respectively, compared to existing models with 0.48 and 0.49, respectively. The distinctive feature of the clusterizer is the use of more accurate input assessments, determined by the principle of interaction and linear scaling. The proposed algorithm involves using a modified chi-square distance metric, which includes assessments of state security indices. A notable feature of the proposed approach is the more accurate determination of cluster centers using Kohonen self-organizing maps. This helps solve the task of analyzing and improving the accuracy of predicting air threat levels. The results are explained by the use of more accurate input assessments and a well-chosen distance metric between clusters in combination with Kohonen self-organizing maps. In practice, the results could be used for analyzing air danger levels by a ground-based platform