Project Title

AI for Diabetes Care

Academic department under which the project should be listed

CCSE - Information Technology

Faculty Sponsor Name

Liang Zhao

Abstract (300 words maximum)

Diabetes is a disease that affects over 400 million people in the world, while the disease is not curable it is still necessary for a patient to be monitoring their glucose concentration levels to live a healthy life. Previous technologies have been made to create measurements that are invasive to the patient by pricking their skin to collect blood, technologies such as the HbA1c and glucometer. These procedures cause discomfort to the patient and can have outdated data be used for the patient, resulting in undetected-cases and other health-complications. Overcoming this problem, is with a non-invasive ultra-wideband blood glucose measurement system that allows for no blood to be collected. This allows patient’s data to be collected and presented regularly using a graphical user interface, but most importantly this data can be entered in an artificial intelligence to detect if the patient is diabetic. The machine learning model chosen to detect diabetes in a patient was a logistic regression model that is best for a binary dependent model, such as being positive or negative with diabetes. For the most accurate model, only real-world data from “Pima Indians Diabetes Database” was used to train the machine learning model, and included data of patient's glucose levels, pregnancies, insulin levels, BMI, skin thickness, age, blood pressure, diabetes pedigree function, and the outcome of if the patient has diabetes. The training and testing of the machine learning model resulted in a 79% accuracy rate to data presented. This system will help provides patients with longer lives and allows undetected cases of diabetes to be reduced.

Project Type

Oral Presentation (15-min time slots)

How will this be presented?

Yes, synchronously via Teams

This document is currently not available here.

Share

COinS
 

AI for Diabetes Care

Diabetes is a disease that affects over 400 million people in the world, while the disease is not curable it is still necessary for a patient to be monitoring their glucose concentration levels to live a healthy life. Previous technologies have been made to create measurements that are invasive to the patient by pricking their skin to collect blood, technologies such as the HbA1c and glucometer. These procedures cause discomfort to the patient and can have outdated data be used for the patient, resulting in undetected-cases and other health-complications. Overcoming this problem, is with a non-invasive ultra-wideband blood glucose measurement system that allows for no blood to be collected. This allows patient’s data to be collected and presented regularly using a graphical user interface, but most importantly this data can be entered in an artificial intelligence to detect if the patient is diabetic. The machine learning model chosen to detect diabetes in a patient was a logistic regression model that is best for a binary dependent model, such as being positive or negative with diabetes. For the most accurate model, only real-world data from “Pima Indians Diabetes Database” was used to train the machine learning model, and included data of patient's glucose levels, pregnancies, insulin levels, BMI, skin thickness, age, blood pressure, diabetes pedigree function, and the outcome of if the patient has diabetes. The training and testing of the machine learning model resulted in a 79% accuracy rate to data presented. This system will help provides patients with longer lives and allows undetected cases of diabetes to be reduced.

blog comments powered by Disqus