The rapid proliferation of Internet of Things (IoT) devices has led to a growing number of security threats, highlighting the limitations of traditional Intrusion Detection Systems (IDS) in dynamic and resource-constrained IoT environments. Conventional IDS often rely on single classifiers, which struggle to balance accuracy, speed, and adaptability. In response, this study introduces a novel hybrid ensemble learning framework that integrates four powerful tree-based algorithms: Random Forest, LightGBM, CatBoost, and XGBoost. Each model brings unique strengths—Random Forest’s robustness to overfitting, LightGBM’s speed, CatBoost’s handling of categorical features, and XGBoost’s high predictive power. These models are combined using a soft voting mechanism, where the final prediction is determined by the weighted probabilities from each classifier. This design aims to enhance detection accuracy while preserving real-time responsiveness. The proposed method is evaluated using two benchmark intrusion detection datasets: UNSW-NB15 and CICIDS2017. Experimental results show that the hybrid approach consistently outperforms individual models and state-of-the-art techniques, achieving up to 99.98% accuracy, with notable improvements in precision, recall, and F1-score.