Athletes need precise, individualized macronutrient planning according to their physiological needs, intensities and recovery phases of training. However, such planning using conventional dietary approaches often rely on static guidelines that do not account for the temporal changes in energy expenditure and individual metabolic variability. Addressing this limitation, this research developed an adaptive and data-driven nutrition forecasting using artificial intelligence and machine learning models. This research performs an extensive evaluation of different machine learningalgorithms, such as linear regression, random forest, XGBoost, LightGBM and long short-term memory (LSTM), to forecast the carbohydrate, protein and fat needs of athletes. It benchmarks these models on their performance concerning accuracy and generalization. The LSTM model demonstrated the highest predictive performance, while other models showed enhanced interpretability. To use the strengths of such models, this research suggests a hybrid framework that integrates LSTM with feature optimization for the extraction of temporal features. These results demonstrate that hybrid architectures with the benefits of AI hold promise for personalized and real-time nutrition recommendations by enhancing performance, recovery and precision nutrition, merging both methodological and practical experiences in dietary management driven by artificial intelligence.