IoT Threat Detection in Information Systems by ChaCha20-Poly1305-Protected Hybrid Light-GBM–Genetic Algorithm Framework

The growing of IOT, it proposes a dataset with security framework built into the hybrid model for IoT-enabled information systems based on Light-GBM-based intelligent threat detection, Genetic Algorithm feature optimization, and ChaCha20-Poly1305 authenticated encryption for output protection. The overall framework proposed is composed of multi-source IoT data acquisition and pre-processing alongside the temporal and contextual feature transformation. Specifically, first Light-GBM is employed to rank candidate features and then a Genetic Algorithm applies to find an optimized feature subset based on a fitness function which consists of several measures including F1-score, recall and compactness of the subset. The selected optimal subset is then used to train the final Light-GBM detection model, while ChaCha20-Poly1305 ensures that security-related outputs are protected against tampering and unauthorized disclosure. Experimental evaluation performed on aggregated IoT datasets proved that this model reached an accuracy of 99.54%, precision equal to 97.31%, recall equal to 99.52%, F1-score turned out to be equal to 98.41% and ROC-AUC measured 99.953%. The new model also had 2,204 false negatives compared with the planned Light-GBM configuration — but only 992 in comparison to the prototype Light-GBM model that we finally evolved — implying a sharply improved detection sensitivity. The proposed framework provides a more integrated security architecture by combining optimized detection with lightweight authenticated protection despite the overall classification metrics being slightly higher for Decision Tree and Random Forest. These results prove that our proposed approach is a pragmatic and efficient enhancement for the IoT threat detection and securing the outputs of information systems.