Comparative Evaluation of Machine Learning Models for UAV Network Performance Identification in Dynamic Environments

Abstract

The rapid integration of Unmanned Aerial Vehicles (UAVs) into critical applications such as disaster management, logistics, and communication networks has brought forth significant challenges in optimizing their performance under dynamic and unpredictable conditions. This study addresses these challenges by systematically evaluating the predictive capabilities of multiple machine learning models for UAV network performance identification. Models including RandomForest, GradientBoosting, Support Vector Classifier (SVC), Multi-Layer Perceptron (MLP), AdaBoost, ExtraTrees, LogisticRegression, and DecisionTree were analyzed using comprehensive metrics such as average accuracy, macro F1-score, macro precision, and macro recall. The results demonstrated the superiority of ensemble methods, with ExtraTrees achieving the highest performance across all metrics, including an accuracy of 0.9941. Other ensemble models, such as RandomForest and GradientBoosting, also showcased strong results, emphasizing their reliability in handling complex UAV datasets. In contrast, non-ensemble approaches such as LogisticRegression and MLP exhibited comparatively lower performance, suggesting their limitations in generalization under dynamic conditions. Preprocessing techniques, including SMOTE for addressing class imbalances, were applied to enhance model reliability. This research highlights the importance of ensemble learning techniques in achieving robust and balanced UAV performance predictions. The findings provide actionable insights into model selection and optimization strategies, bridging the gap between theoretical advancements and real-world UAV deployment. The proposed methodology and results have impact for advancing UAV technologies in critical, network performance-sensitive applications.