Smart cities rely heavily on Internet of Things (IoT) technology, which enables automation services through interconnected IoT devices. However, the widespread use of IoT applications in smart cities has resulted in security and privacy concerns that must be addressed to protect sensitive data. To safeguard smart cities from cyber attacks, learning theory-based automated attack-detection methods must be adopted. Various techniques have been proposed in the literature to create effective models for identifying IoT attacks. However, the majority of IoT detection algorithms have focused on only a few types of IoT attacks, and most IoT threat detection systems have used black-box deep learning models that lack interpretability to support their forecasts. This research aims to detect several types of large-scale attacks on IoT devices using the Extreme Gradient Boosting (XG-Boost) classifier and Explainable Artificial Intelligence (XAI) approaches. The proposed method not only improves the model’s performance but also increases trust in the model. The results of the experimental study on the IOTD20 dataset and XAI evaluation of each feature’s contribution to the model demonstrate that the proposed model can efficiently identify malicious attacks and threats, reducing IoT cybersecurity threats in smart cities.