Pima Indians Diabetes classifier

Inspiration

Diabetes is a major global health concern influenced by a combination of genetic, environmental, and lifestyle factors. Pima Indian women experience disproportionately high rates of diabetes, which inspired us to investigate whether health measurements could be used to predict diabetes within this population.

What It Does

This project analyzes the Pima Indians Diabetes dataset and builds a machine-learning model that classifies whether a patient is likely to have diabetes.

The model uses health-related features such as glucose level, blood pressure, skin thickness, insulin level, BMI, and age. It also includes visualizations and performance metrics to explain the model’s results and limitations.

How We Built It

We completed the project in R using Jupyter Notebook. First, we cleaned the dataset, converted the diabetes outcome into a categorical variable, and divided the data into training and testing sets.

We then standardized the predictor variables and trained a k-nearest neighbors classification model using the tidymodels framework. To select an appropriate number of neighbors, we tested multiple values of k using five-fold cross-validation and visualized their estimated accuracy.

Finally, we evaluated the model on unseen testing data using accuracy, precision, recall, and a confusion matrix. The final model achieved approximately 74% accuracy, but its low recall demonstrated that accuracy alone is insufficient—especially in a health-related application where missed positive cases can have serious consequences.

Built With

R, Jupyter Notebook, tidyverse, tidymodels, ggplot2, kknn, k-nearest neighbors, five-fold cross-validation, and the Pima Indians Diabetes dataset.

This project is an educational data-science analysis and should not be used as a medical diagnostic tool.

GitHub repository