Аннотация. The rapid expansion of the Internet of Things (IoT) has broadened the digital attack surface, exposing network infrastructures to advanced threats such as ARP spoofing. Traditional detection systems, often relying on rule-based or signature-driven methods, struggle to adapt to the dynamic and heterogeneous nature of IoT environments. To address this limitation, this paper proposes a detailed evaluation of several detection models for Address Resolution Protocol (ARP) spoofing attacks within resource-limited Internet of Things (IoT) environments. Focusing on both detection accuracy and computational efficiency, six models—comprising traditional classifiers and hybrid ensembles—are compared using metrics such as accuracy, precision, recall, F1-score, training/testing time, and memory usage. The proposed OSK hybrid model, which integrates Support Vector Machine and k-Nearest Neighbors, achieves the highest detection performance while maintaining moderate resource consumption. The findings reveal important trade-offs between detection capability and system constraints, emphasizing the effectiveness of optimized hybrid solutions for real-time implementation in IoT networks. This work contributes valuable guidance for developing efficient and reliable ARP spoofing detection frameworks tailored to the demands of IoT devices.