Modified logistic regression algorithm for accurate determination of heart beats from noisy passive RFID tag data

Passive RFID tags provide a promising way to create wireless and battery-free heart rate monitors. However, the reliability of these tags is limited in the presence of common noise sources in their environment. In this paper, we propose an algorithm to improve the beat detection for RFID based heart rate monitors in noisy environments. To achieve this, a logistic regression model is first employed to determine data points that have a very high probability of being actual heart beats. These data points are then used as features to remove the ambiguity in detection of other heart beats. The algorithm is trained using features from a single heart rate measurement and the obtained parameters are used for determining various other heart rates. Using our algorithm, we achieve an F1-score of 0.98 for correct heart beat detection, and completely eliminate an error of over 75% in mean heart rate calculation.