Fog computing, which extends cloud computing to the edge of the network, presents new opportunities for low-latency processing and decentralized decision-making. However, the distributed nature of fog networks also introduces significant security challenges, including the risk of malicious node behavior, such as ballot stuffing in voting or data-sharing systems. In this paper, we propose a novel approach for detecting malicious nodes in fog computing networks using a combination of K-Means clustering and Dynamic Threshold-Based Outlier Detection. Our method leverages trust scores calculated from node reputation, behavior, and interaction frequency, and applies clustering to group nodes based on their trustworthiness. Nodes that deviate significantly from their respective cluster centroids are flagged as suspicious and potentially malicious. We evaluate our approach through extensive experiments, comparing it against baseline methods, including threshold-based detection and random selection. The results demonstrate that our approach achieves higher precision, recall, and F1-score, with a lower false positive rate, while maintaining reasonable computational efficiency. The proposed method offers a scalable and robust solution for real-time malicious node detection in fog networks and can be applied to a wide range of fog-based applications, including distributed voting systems, IoT networks, and collaborative decision-making frameworks.