<![CDATA[Interpreting physiological health data and self-reported nutrition records often poses a computational challenge for non-expert users. Therefore, this study aims to conduct a comparative analysis of fitness level classification modeling and predictive calorie regression evaluation. Both are integrated into a unified health tracking ecosystem called FitTrack AI. The performance of the Random Forest, XGBoost, and Support Vector Machine (SVM) algorithms was comprehensively compared for multi-class classification. Meanwhile, calorie burn estimates were evaluated using the Random Forest Regressor. As a holistic system, this ecosystem is also supported by body weight projection analysis (Linear Regression), dietary pattern mining (Apriori), and an automated logging interface based on a Large Language Model (Groq API). Test results show that XGBoost is the best classification model, with an accuracy rate of 76.37%, outperforming other algorithms. In the calorie prediction regression test, the model achieved highly accurate performance with a coefficient of determination (R²) of 0.996. In terms of ecosystem functionality, the interactive virtual assistant (FitBot) recorded a 90.0% success rate in executing tool calls for data entry and achieved a System Usability Scale (SUS) score of 90.1 (Very Good category). Overall, this multi-model analytical approach has proven to be robust and effective in translating the complexity of biological data into comprehensive and personalized digital health insights.]]>
