A Cognitive AI Model Using Deep & Cross Network for Personalized Gamified Recommendations to Promote Healthy Eating Decisions Among University Students

Unhealthy eating habits among university students represent a critical public health challenge, adversely affecting physical health, mental well-being, and academic performance. While existing solutions leverage mobile apps and wearables, they lack robust cognitive artificial intelligence for personalized interventions and fail to sustain engagement through gamification. Previous efforts have employed deep learning models such as LSTM for dietary event detection and gamified frameworks, yet they neglect the integration of AI precision with long-term behavioral incentives. Current tools exhibit limited personalization, poor scalability, and insufficient validation in real-world contexts, particularly for diverse student populations. Motivated by these gaps, this study proposes a cognitive AI framework using a Deep and Cross Network integrated into a gamified mobile application to predict nutritional status and deliver tailored dietary challenges. The DCN model was trained on data from 5,000 Peruvian students, incorporating behavioral, demographic, and biometric variables, and validated against baseline models using stratified 80-20 splits, five-fold cross-validation, and metrics including Precision, Recall, F1-Score, and AUC-ROC. The DCN achieved superior precision of 92.89% and AUC-ROC of 94.02%, outperforming BNN and NCF by modeling nonlinear feature interactions such as BMI and sleep quality. This research advances human-AI interaction by synergizing cognitive modeling with gamification, demonstrating enhanced user engagement and offering a replicable framework for sustainable dietary behavior change.