Snoring has a significant effect on human health, and it is a prevalent issue worldwide. Therefore, this study aims to establish a machine learning algorithm to detect snoring. The acoustic features of snoring are extracted based on Mel-Frequency Cepstral Coefficients (MFCCs) and Zero-Crossing Rate (ZCR). The extracted features are inputs for the cascaded model, where the first layer, the Random Forest (RF) model, is used to evaluate the features’ importance and extract the most important features. The refined feature set and the RF’s prediction result are then put into the second layer, the Support Vector Machine (SVM) classifier with a Radial Basis Function (RBF) kernel, where the SVM learns when the RF’s prediction results are trustworthy, while maintaining the ability to make independent decisions using its advantages in modeling non-linear signals. Therefore, this cascaded model structure fully utilizes the strengths of both the RF and the SVM. The accuracy of the model reached above 95%, which is comparable to current state-of-the-art methods, demonstrating the effectiveness of the proposed structure. More research can be done in the future to further improve its accuracy in real-world scenarios with larger, multimodal datasets.