Ilham A.E. Zaeni, Andriana Kusuma Dewi, Putra Wisnu Agung Sucipto, Muhammad Khusairi Osman
This study introduces a spectral feature extraction framework based on baseline normalization for fatigue classification using electrocardiogram (ECG) signals. ECG signals are transformed into frequency-domain features through the Fast Fourier Transform (FFT), after which baseline-normalization based on subtraction and division techniques are applied. These processed features are used to train a Decision Tree classifier to classify the fatigue level into low, medium, and high fatigue. Experiments were conducted on the FatigueSet dataset, which contains ECG signals from 12 subject who participated into experiment to induce cognitive workload. Results highlight that baseline subtraction yields superior outcomes, with 86% accuracy compared to 65% from division and 53% without normalization. The baseline subtraction method also gives a balance result on precision, recall, and F1-score with score 89%, 86%, and 86%, respectively. Moreover, the decision tree generated interpretable rules based on frequency bins, confirming that baseline normalization improves both model performance and clarity for practical wearable applications. © 2025 IEEE.
Universitas Negeri Malang, Dept. of Electrical Engineering and Informatics, Malang, Indonesia
