The spread of Internet of Things (IoT) devices encompasses enormous security flaws in their network. The integration of Intrusion Detection Systems (IDS) that rely on adaptive machine learning (AML) into IoT networks serves to address the distinct security issues that arise from the variety and widespread use of IoT devices. In this paper, a hybrid AML-based IDS using deep learning models, such as LSTM and CNN + GRU to detect anomalies and threats such as DDoS, DoS, and U2R has been adopted to evaluate its anomaly detection capability in a variety of datasets, such as UNSW_NB15, CICIDS2017, NID, and TON_IoT. When tested on these datasets, the accuracy of 98.9%, 99.5%, 98.7%, and 99.9% are obtained on the UNSW_NB15, CICIDS2017, NID, and TON_IoT datasets respectively by the Max Voting mechanism. Furthermore, the MSE and R2_score for the hybrid model for all the datasets were found to be ~ 4 × 10–5 and ~ 0.99 respectively, with ROC-AUC plots of ~ 0.9, which further proves its supremacy. The study in addition highlights the supremacy of AML-based IDS in protecting IoT networks and the potential of ML methods to improve cybersecurity through a comparative performance analysis with related works. The results point out the need for IDS to strike a balance between reducing false alarms and having a high detection rate with the goal of protecting IoT devices.