Intelligent application for calorie calculation and personalized physical activity recommendations for patients with obesity

Abstract

Obesity management increasingly requires digital tools that move beyond generic calorie tracking toward context-aware and clinically informed decision support. This study presents an intelligent mobile application for personalized calorie calculation and physical activity recommendation in adults with obesity, integrating a random forest (RF) risk model, a Flutter-based interface, and a secure cloud backend. The study used 768 anonymized clinical records from Peruvian healthcare institutions and a four-week pilot evaluation with 52 adult users. After preprocessing, class balancing, and five-fold cross-validation, RF was retained because it offered the best compromise between classification performance and interpretability when compared with alternative models explored during development. The final model achieved 74.03% accuracy, 68.29% recall, 70.98% F1-score, and 0.77 area under the curve (AUC) on the held-out test set. In pilot use, weekly exercise frequency increased by 20%, mean glucose decreased by 5%, and usability was rated highly (system usability scale (SUS)=84.7). The main contribution of this work is the integration of locally contextualized data, explainable risk estimation, personalized behavioral feedback, and scalable mobile health deployment for obesity self-management in resource-constrained settings.