Network security in Internet of Things (IoT) environments poses challenges due to the heterogeneous nature of devices, resource constraints, and the scale of data generation. This paper proposes a novel Support Vector Machine (SVM) based approach for detecting anomalies and security threats in IoT networks. The method we propose implements a two-stage detection framework that uses features which are optimized for IoT traffic patterns with an adaptive SVM classifier. The model was evaluated and trained on a dataset which consists of 1.2 million network flows collected from 2,500 IoT devices across industries and smart home environments. Our approach achieves detection accuracy of 97.3% with a false positive rate of 0.8%, which outperforms traditional anomaly detection methods by an average of 12%. Our proposed method performs reasonably well in identifying unknown cyberattacks attacks while maintaining computational efficiency which suits resource-constrained IoT environments. Our implementation also reduces the feature extraction overhead by 45% in real time IoT devices. Our findings show that anomaly detection based on SVM can be effectively adapted for IoT security while addressing the unique constraints of the field.