Alper A., Sağır N., Hasırcı Tuğcu Z.
9th International Artificial Intelligence and Data Processing Symposium (IDAP), Malatya, Türkiye, 6 - 07 Eylül 2025, ss.1-6, (Tam Metin Bildiri)
Özet
This study aims to enhance the reliability of Vehicle-to-Vehicle (V2V) communication in Intelligent Transportation Systems (ITS), particularly under conditions where signal degradation is caused by environmental obstacles. A lightweight and interpretable machine learning framework is proposed that operates solely on physical-layer signal features-namely, path loss and inter-vehicle distance-without relying on visual or contextual sensor data. Measurement campaigns conducted in Trabzon (urban) and Gümüşhane (rural/mountainous) regions of Türkiye provided a balanced and representative dataset involving three dominant obstacle types: vehicles, buildings, and trees. Using this dataset, four supervised classification models-KNearest Neighbors (KNN), Random Forest (RF), Support Vector Machine (SVM), and Logistic Regression (LR)-were implemented, hyperparameter-optimized, and thoroughly evaluated. The results demonstrate that KNN (Accuracy: 98.81%, F1Score: 0.9877) and RF (Accuracy: 99.54%, F1-Score: 0.9952) significantly outperform linear and kernel-based alternatives. Their high generalization capabilities against nonlinear obstacle patterns are further confirmed by confusion matrix and ROC curve analyses. The proposed system is computationally efficient, scalable, and suitable for integration into adaptive power control or routing strategies in ITS applications. This work introduces a novel approach to V 2 V communication by linking physical-layer signal behavior with high-level environmental awareness through RF-based machine learning classification.