<p>The Internet of Things (IoT) is a rapidly growing technology that has become an integral part of various industries, from smart homes to healthcare and beyond. However, IoT devices are highly vulnerable to malicious attacks, making effective intrusion detection a critical component for ensuring their security. This paper presents an intelligent intrusion detection framework that integrates Decisive Red Fox (DRF) and Grey Wolf Optimization (GWO) for feature extraction, combined with a hybrid deep learning (DL) and machine learning (ML) classification approach. The proposed framework first applies DRF-GWO to select a discriminative subset of features, which enhances the representation of attack patterns and improves the data quality for the subsequent classifier. In the classification phase, Convolutional Neural Networks (CNN) are employed to extract spatial patterns, followed by Bidirectional Long Short-Term Memory (BiLSTM) networks to capture temporal dependencies in network traffic. The final classification is performed using Extreme Gradient Boosting (XGBoost), which enhances accuracy and robustness. Experimental results on the Ton-IoT and Bot-IoT datasets demonstrate that the feature selection stage, while adding an initial time cost, leads to a more streamlined and effective classification model. The proposed system achieves remarkable accuracies of 98.15% and 98.67%, respectively, significantly outperforming traditional intrusion detection methods. By integrating DRF-GWO feature extraction with a CNN-BiLSTM-XGBoost classifier, the framework substantially improves detection rates and reduces false positives, making it a highly effective solution for IoT security applications where accuracy is paramount.</p>