Ilham A.E. Zaeni, Agung Nugroho
This paper uses a decision tree classifier and Fast Fourier Transform (FFT)-based feature extraction to make a system that can detect when someone is snoring. The proposed framework solves two problems with automated snoring detection: being sensitive to background noise and not being able to work quickly in real time. A Butterworth bandpass filter (100 ~Hz-4 kHz) is used to clean up raw audio signals and keep the harmonics of snoring. Using Pearson correlation, the spectral features extracted with FFT are improved to keep the top ten most discriminative frequency bins. The detection system was built using a Kaggle dataset with 1, 2 0 0 samples that anyone can access. We use the hold-out method with a training dataset that is 70% of the total and a testing dataset that is 30% of the total to check the model. The decision tree model is good for environments with limited resources because it has an accuracy of 73.3% and keeps a balance between precision and recall. Because it is easy to understand, efficient, and simple, the system can be used in health monitoring devices that people wear. © 2025 IEEE.
State University of Malang, Dept. of Electrical Engineering and Informatics, Malang, Indonesia
