JMIR Publications

ABSTRACT

Background:

Diabetes is a chronic disease with a high global prevalence, increasing from 200 million in 1990 to 830 million in 2022, with a higher burden in low/middle-income regions and high mortality in Mexico and Veracruz [1]. These inequalities limit access to treatments and nutritional education, which is why technological solutions such as interactive kiosks based on artificial intelligence are required to contribute to the nutritional management of people with diabetes in marginalized communities.

Objective:

To design and evaluate an artificial intelligence (AI)-based interactive kiosk that generates culturally relevant and personalized meal plans for people with diabetes in underserved communities.

Methods:

A low-cost prototype was developed using a local food database and a multilayer perceptron trained on synthetic data based on national clinical guidelines. Performance was tested through an experimental evaluation that measured: (i) recommendation accuracy against idealized plans (accuracy, precision, sensitivity, and F1); (ii) throughput, recording response time with 1‑50 simultaneous requests; and (iii) usability, using heuristic evaluation and the System Usability Scale (SUS).

Results:

The smart kiosk was experimentally evaluated in three dimensions: nutritional recommendations, system efficiency, and usability. The AI model achieved an overall accuracy of 88% (precision 85%, sensitivity 91%, F1 88%) without overfitting, outperforming the traditional method by 28 points. Response time ranged from 1.2 to 4 s depending on the load, while maintaining resource stability. The interface achieved a success rate of 98.3% and a SUS score of ~90/100.

Conclusions:

The AI-based kiosk demonstrated technical feasibility, high response speed, and excellent usability, significantly improving nutritional management. It achieved 88% concordance with nutritionist plans versus 60% for rule-based systems, thanks to its dynamic macronutrient adjustment and incorporation of local foods. It responds in less than two seconds, runs on low-cost hardware, and achieved a SUS score close to 90, ensuring a seamless and accessible experience.