This paper presents a method for detecting counterfeit accelerometer integrated circuits (ICs) by analyzing their noise signatures. The method utilizes Gaussian Mixture Model (GMM) clustering, a probabilistic machine learning algorithm, to group accelerometers based on their Allan Variance characteristics and identify anomalous clusters that may contain counterfeit components. The potential of this technique is explored using experimental data from two types of consumer-level accelerometers. Presented results suggest that the proposed approach can accurately differentiate between accelerometer parts with high probability, based solely on noise properties of the parts. The proposed method can be included in the manufacturing process flow as an additional non-destructive testing method to probabilistically identify non-authentic parts. The cut-off threshold for detection of counterfeit parts is a user-chosen probability value. Lastly, the benefits and drawbacks of the method, as well as potential avenues for future research, are discussed.