This paper intends to prove a novel bidirectional long short-term memory (LSTM) model designed for real-time traffic pattern prediction and anomaly detection in Internet of Things (IoT) applications. The introduced model combines unsupervised and supervised learning techniques to achieve two main goals: predicting normal traffic behavior and identifying malicious activities. Leveraging a bidirectional LSTM network, the model demonstrates exceptional performance in detecting both known and unknown threats. When evaluated on an intrusion detection dataset, the model achieves 98% to 99% accuracy in threat detection while maintaining low computational cost. This efficiency makes the model particularly suitable for deployment on resource-constrained IoT devices. The model’s hybrid architecture, which combines prediction and anomaly detection capabilities, provides a comprehensive and effective approach to enhancing IoT security.