A Scalable Hybrid RF-BiLSTM Framework for Reliable IoT Traffic Threat Detection via Feature Selection and Temporal Pattern Recognition

In this research, we addressed the recurring challenges of securing IoT networks against emerging cyber security threats. Taking advantage of the complementary strengths of Random Forest (RF) for feature selection and Bidirectional Long Short-Term Memory (BiLSTM) networks for sequential learning; we developed a novel Hybrid RF-BiLSTM model that combines feature level insights with temporal pattern recognition to provide a reliable solution for IoT traffic threats. We conducted extensive experiments with Aposemat IoT-23 dataset, where we used equal volumes of benign and malicious traffic samples leading to balanced evaluation. Furthermore, the Hybrid RF-BiLSTM model achieved a performance of 99.87%, while the Random Forest and BiLSTM performance were 99.37% and 93.32%, respectively, demonstrating the power of the hybrid approach over individual ones. The analysis gave more details about the model's performance, showing the confusion matrix and calculating the performance metrics that substantiates the model's reliability to minimize false positive and false negatives while also achieving high precision and recall. It shows that how well the integration of feature selection and sequential learning works for IoT cyber security. This Hybrid RF-BiLSTM approach lays a scalable and practical framework for real-world IoT security problems and a stepping stone for future studies in hybrid ML models for anomaly detection and threat analysis.