Abstract: This study proposes a classification model based on a decision tree trained with features extracted from heart rate variability signals, specifically in the frequency domain, using low frequency and high frequency bands derived from RR intervals. Through the extraction of averages from non-overlapping windows for 24-hour RR intervals, representative features were generated for each patient. The model achieved an accuracy of > 70%, demonstrating that HRV signals in the frequency domain are useful for classifying and stratifying patients with Chagas disease
Method: In this study, data from 83 control group patients, 102 ch1 group patients and 107 ch2 group patients were analyzed using time series of RR intervals. The signals were transformed to the frequency domain using Fourier transform to extract the LF and HF bands. Twenty-four features per patient were extracted from non-overlapping windows of 12 samples. Additionally, Euclidean distances between the averages of values for each patient were calculated, and adjustments to the averages were made to reduce the impact of standard deviation and outliers. A decision tree model was trained with 80% of the data and evaluated for accuracy on the remaining 20% to classify patients
Results: The decision tree model achieved an accuracy greater than > 70% in the test set using only features extracted from HF and LF, suggesting that HRV analysis in the frequency domain is effective for classifying patients
Conclusions: This study presents an effective model for stratifying patients with Chagas disease using features extracted from RR interval signals in the frequency domain. The high accuracy achieved (greater than 70%) demonstrates that LF and HF bands contain valuable information for classifying physiological states. These findings suggest that HRV analysis could become a useful tool for clinical evaluation of patients with Chagas disease, improving the ability to detect and classify more severe cases