Obstructive sleep apnea (OSA) is a common sleep disorder characterized by repeated interruptions in breathing due to partial or complete blockage of the upper airway. It is often associated with loud snoring, which serves as an important marker for diagnosing the condition. OSA can lead to severe health issues, such as hypertension, stroke, and cardiovascular diseases, if left untreated. Snoring sounds are known to carry valuable acoustic information about the upper airway’s condition, making them useful for non-invasive OSA detection. In this study, we explored the strategy to analyze snoring sounds to detect OSA. We were able to classify different apnea events by extracting pertinent characteristics from tracheal snoring signals using publicly available datasets in EDF format. In order to examine airflow patterns and identify apnea episodes, we derived Mel Frequency Cepstral Coefficients (MFCC) features from tracheal snoring signals. Furthermore, to assess how well these features performed in categorizing apnea events, machine learning techniques including Random Forest, SVM, K-NN, decision tree, and XG-Boost were exploited for classification. With an average precision of 0.81 and an F1-score of 0.62 for three patients and an average precision of 0.83 and an F1-score of 0.67 for eleven patients, the Random Forest classifier performed the best. While Naive Bayes and Logistic Regression scored poorly because they were unable to identify the non-linear relationships in the data, other models such as SVM and XG-Boost had potential but had trouble with class overlaps. Our results show that while other models need to be further improved to handle the complexity of the classification problem, tree-based models, Random Forest in particular, are well-suited for OSA detection using snoring signals.