Logo: to the web site of Uppsala University

Breathing is more than a rhythmic mechanical process. It is also connected to brain processes that regulate cognitive and emotional states. Previous studies have shown that respiratory rhythm can influence neural activity, attention, memory, and emotional processing. This means that breathing patterns may reflect both physiological processes and a person’s internal state. This raises the question of whether breathing signals could contain useful information for identifying individuals and predicting physiological traits. This thesis explores the use of machine learning to predict sex and age, and to identify participants through physiological fingerprinting, using resting-state respiratory and cardiovascular signals. The experiments validate the feasibility of these tasks on a new dataset combining nasal and oral breathing with cardiovascular signals, which has not been done before. Recordings include ECG, fingerpulse, nasal airflow measured with a cannula, and oral airflow measured with a thermopod, collected from 112 healthy  participants. A one-dimensional convolutional neural network was used for sex and age prediction, formulated as both classification and regression tasks, while contrastive learning was used for fingerprinting. The results show that physiological signals contain useful information for sex prediction and participant identification, while age prediction did not improve over the mean baseline. The cannula signal was the most informative individual signal for sex classification, achieving ROC-AUC values of up to 0.74–0.75, while ECG dominated the fingerprinting task, with rank-1 identification accuracy reaching approximately 97% for nasal breathing when ECG was combined with cannula, and approximately 95% for oral breathing using ECG alone or combined with fingerpulse. This study concludes that breathing contains unique information that can contribute to participant identification, although cardiac signals are still the most relevant for predictions. Since respiration is closely connected to cognitive and emotional processes, these findings may have broader implications for non-invasive physiological monitoring. They may also help in understanding how individual differences in physiology and internal state can appear in breathing patterns.