Desarrollo de una aplicación para el diagnóstico de diabetes tipo 2 utilizando aprendizaje máquina

Abstract

Diabetes is a chronic, long-term disease that poses a significant challenge to global public health. The American Diabetes Association (ADA) classifies it as a growing problem due to its association with severe complications that affect patients' quality of life and life expectancy. In 2022, an estimated 463 million people worldwide were diagnosed with type 2 diabetes—a staggering figure that highlights the urgency of implementing innovative strategies for its management and diagnosis. This disease significantly increases the risk of chronic complications such as cardiovascular diseases, strokes, blindness, and amputations, generating a negative impact on people’s health and quality of life. In this context, machine learning is presented as a powerful tool in the diagnosis of diseases. Various studies have demonstrated its effectiveness not only in detecting diabetes but also in diagnosing cardiovascular diseases, breast cancer, COVID-19, and more. This research focuses on developing a web application for the diagnosis of type 2 diabetes using machine learning, guided by the CRISP-DM methodology for the creation and evaluation of classification models. The web application development followed best practices from the XP methodology. The data used was collected from 961 patients’ observations at the Postural Health and Wellness Clinic in Quito- Ecuador, between July and December 2023, within an age range of 30 to 60 years. Among the evaluated models, Random Forest stood out as the most accurate, achieving a 99.46% accuracy and an AUC of 99.98%. The main contribution of this research is to provide an artificial-intelligence-based tool that facilitates healthcare professionals and patients the early diagnose of type 2 diabetes. This solution not only reduces costs associated with diagnostics, but it improves accessibility, and provides support specifically to those who live in rural areas.