Estimation of daily energy requirements using a hybrid artificial intelligence model

Accurately estimating energy requirements is critical for individuals to maintain a healthy life. Traditional methods may be time-consuming, complex, low in accuracy, and costly, thus creating a need for new approaches. This study explores the applicability of hybrid artificial intelligence models for calculating daily energy requirements based on individuals' anthropometric measurements and demographic data. The study's primary goal is to develop a model that offers a reliable and practical solution with higher accuracy than existing methods while ensuring ease of use in field settings. This study used data collected from volunteer individuals at Sakarya University between September 2023 and February 2024. Anthropometric measurements were performed by a bioelectrical impedance analysis (BIA) device, and demographic data were obtained through face-to-face surveys. Eighty-seven features were analyzed using the Spearman feature selection algorithm, and these were utilized to estimate energy requirements. Based on collaborative hybridization, the hybrid artificial intelligence model integrates three distinct Gaussian Process Regression (GPR) models using squared exponential, rational quadratic, and Matern52 kernels. These models were structured based on gender, and performance evaluation was carried out using criteria such as MAPE, MAD, MSE, R, and R². The best model performance in males was achieved at level 10 with 100% R², while the highest accuracy in females was observed at level 15. To increase model simplicity, the PCA method was applied; however, a decrease in performance was detected (R = 0.48, R² = 0.23). The accuracy of the artificial intelligence models proposed in this study was significantly higher than that of traditional formulas commonly preferred in the current literature. Hybrid artificial intelligence models offer practicality, accuracy, and flexibility in estimating energy requirements. Gender-based modeling has enhanced prediction performance while providing more reliable results by accounting for individual differences. This approach holds significant potential for advancing health and nutritional sciences.