Pavan kalyan reddy Kamasaani

This project is to predict whether one has diabetes or not, simply by the responses given by the individual on a health survey, without a blood test or lab work. My dataset was a CDC dataset named BRFSS 2024 containing the answers of nearly 457,670 individuals residing in the United States of America regarding their lifestyle, health, and demographics. I cleaned data, investigated it to identify patterns and trained six different machine learning models to determine which model performed well in identifying diabetic patients. The models continued to predict the type no diabetes to all so that the accuracy was an invalid measure as this was not true to the majority of the dataset. Therefore, rather than aiming at accuracy, I aimed at ensuring that the model captures as many cases of the diabetics as possible. With testing and tuning, I determined that XGBoost was the most suitable model, and by using its decision threshold of 0.50, reducing it to 0.13, the model could catch a minimum of 12 percent of the diabetics and a maximum of 83 percent. The most significant variables that were indicative of diabetes were age, BMI, general health, and physical health. This project has demonstrated that constructing a good model is not only what algorithm you choose, but also the way in which you get to know the issue inside and out and then make the appropriate changes to make it actually work in real life.

none