The rapid integration of the digital and physical realms by an IoT is having significant ramifications for people’s lives. Although technology has made many things easier, it may still be hacked or otherwise compromised. With the proliferation and sophistication of network threats, most companies have begun incorporating IDS into their security framework. Detecting network intrusions (or attacks) has grown more important in recent years, with machine-learning algorithms playing a major role. This research aims to provide a novel paradigm for improving IDS efficiency in IoT contexts via the use of machine learning. This work employed the IOT Weather Dataset, which was split into train and test sets for ML models like LGBM and Extra Trees classification and intrusion detection. The input dataset is significantly imbalanced; to tackle this issue, the ADASYN data balancing approach was applied. According to F1-score, precision, accuracy, and recall, a trial outcome reveals that the suggested approach functions well. The proposed models achieve high accuracy of 97.61, and 96.47%, respectively, in terms of increasing network security in IoT traffic. The experimental assessment shows that the proposed design is more accurate and has better detection rates than the standard methods.