An innovative approach for personalized energy expenditure estimation using physiological measurements and gait characteristics: an artificial intelligence-based study

Abstract Accurate estimation of energy expenditure (EE) plays a critical role in combating physical inactivity and obesity. This study aimed to develop a new personalized EE estimation model that goes beyond standard accelerometer data by integrating physiological measurements, walking characteristics, and advanced artificial intelligence techniques. The study involved 60 healthy volunteers (31 men, 29 women) who performed walking tests on a treadmill at different inclines (0%, 2%, 4%). During these tests, indirect calorimetry, accelerometer, ECG, and 3D gait analysis data were collected. Long Short-Term Memory (LSTM) models were used for activity and incline classification, while various machine learning algorithms such as Artificial Neural Networks (ANN), Random Forest, and Support Vector Machines (SVM) were tested for energy expenditure estimation. The results showed that the hybrid model, which integrates heart rate variability (HRV) and walking characteristics, demonstrated significantly superior prediction performance (R²=0.98) compared to basic models using only anthropometric data and accelerometer information (R²=0.77). These findings demonstrate that integrating physiological and kinematic data into artificial intelligence models significantly increases the accuracy of personalized energy expenditure estimation. This innovative approach has the potential to offer a more robust and personalized method for physical activity tracking in areas such as public health, sports analytics, and rehabilitation.