Abstract—Many crucial domains have already begun integrating the Internet of Things (IoT) to serve people better and make their lives easier. The Internet of Things revolution has reshaped digital services across industries by increasing efficacy, productivity, and cost-effectiveness. The security of IoT devices is still a matter of concern despite many service providers having incorporated or adapted them into core portions of their systems’ operation. Machine learning-based anomaly detection has the potential to design an efficient protection system against various IoT cyberattacks, hence reducing the risk to IoT networks. Many detection methods have been developed, current methods only deal with a subset of cyberattacks and evaluate their efficacy using increasingly antiquated datasets. This paper proposes a smart, efficient, and lightweight model for detecting multiple IoT threats. It is called fine-tuned Random Forest- Support Vector Machine (SVM-RF). To construct a powerful and efficient detection model, our proposed framework uses a shared approach for selecting features to choose the most informative and relevant features. Moreover, we suggested a fine-tuned hybrid learning algorithm to enhance classification for predicting multiple types of IoT assaults during the detection phase. Our testing results demonstrate that our proposed method has a good level of accuracy (97%), precision (97%), recall (96%), and F1−Score (97%) in predicting a variety of IoT threats.