The rapid expansion of IoT devices has increased security risks due to limited resources, diverse protocols, and distributed environments. Traditional intrusion detection systems often fail to achieve high accuracy with low false alarms in such conditions. This study presents a Hybrid-IDPS that combines statistical traffic analysis with machine learning–based anomaly detection and feature enhancement. It integrates rule-based detection for known threats with a CNN and gradient boosting model for improved classification. A tiered edge–fog architecture enables real-time processing with low overhead. Experimental results show 98.7% accuracy and significantly reduced false positives, highlighting its efficiency and scalability for modern IoT applications.