BACKGROUND AND OBJECTIVE: Developing predictive computational models of metabolism using mechanistic approaches is complex and resource intensive. Data-driven models offer a reliable, fast, and continuously updating solution for predictive analytics. Previously, we developed the Personalized Metabolic Avatar (PMA), a gated recurrent unit deep learning model, to forecast personalized weight variations based on macronutrient composition and daily energy balance. This model allows for diet plan simulations and tailored goal setting, empowering individuals with the knowledge to achieve long-lasting healthy lifestyle results. However, the PMA requires adaptation through the collection of individual-specific data. Our objective is to address this limitation by creating a more generalized model that maintains predictive accuracy without the need for individual data measurement. METHODS: We propose the Generalized Metabolic Avatar (GMA) to generalize metabolic predictions for a broader user base by incorporating parameters such as age and gender, thus eliminating the need for individual data measurement and enabling application to individuals who have not been previously analyzed. The GMA's predictive accuracy was assessed in both ideal conditions and real-world scenarios. Comparative evaluations against the PMA were performed to validate the GMA's viability and efficiency. RESULTS: The GMA demonstrated promising predictive accuracy, with an average RMSE of 0.54 ± 0.03 in ideal conditions and 0.92 ± 0.76 in real-world scenarios. Comparative evaluations showed that the GMA maintains comparable accuracy to the PMA (PMA RMSE: 0.42 ± 0.04
GMA RMSE: 0.44 ± 0.17), while significantly reducing computational time (PMA: 12.0 ± 1.22 s
GMA: 0.15 ± 0.11 s). CONCLUSIONS: The GMA offers significant advantages over the PMA by employing a single, scalable model that captures common weight fluctuations through age and gender distinctions. It reduces overfitting and enhances generalizability, achieving comparable accuracy to complex deep learning models. The GMA significantly improves computational efficiency by eliminating the need for individual retraining and maintains robust predictive performance even with limited user-specific data.