Avionic platforms are increasingly reliant on sophisticated onboard software and next-generation communication systems to carry out their operations. However, the expanding connectivity capabilities also introduce security vulnerabilities and the potential for attacks, posing significant challenges to the safety of these systems. In particular, the MIL-STD-1553 military standard, widely used as a serial communication protocol for mission-critical systems, lacks specific design considerations for security. The absence of authentication mechanisms in it allows for the potential impersonation of the Bus Controller (BC) and the injection of fake command words. Such attacks pose significant risks to the safety and integrity of mission-critical systems, highlighting the need for robust security measures. In this paper, we propose an unsupervised machine learning technique for anomaly detection on the MIL-STD-1553 communication bus using real platform data. We introduce a novel voting ensemble technique that combines Nearest Neighbors with thresholding algorithms such as Elliptical Boundary, Median Absolute Deviation, and Z-Score. The evaluation results demonstrate that our solution outperforms existing unsupervised machine learning techniques. The proposed method is very likely to be dependable and useful, as it provides a zero false positive rate.