DiabetesPredictor

Overview

“DiabetesPredictor” is a reproducible machine learning software package developed for the BIO215 Capstone Project. It encapsulates a pre-trained Random Forest classifier derived from Kaggle: the BRFSS 2015 dataset to predict diabetes risk based on clinical health indicators.

Features

• Pre-trained Model: Includes a high-performance Random Forest model (AUC ~ 0.82).
• Automated Preprocessing: Automatically handles factor level mapping (e.g., converting numeric inputs like 1 to Excellent for General Health) to ensure robust inference.

Interactive Web Application

Experience the model in action through our user-friendly Shiny interface. You can explore the dataset, visualize distributions, and perform real-time risk assessments.

👉 Live Demo: Launch DiabetesPredictor App

Installation

You can install the development version of DiabetesPredictor from GitHub with:

# install package
remotes::install_github("colinzyang/DiabetesPredictor")

Usage Example

This package comes with built-in example data to demonstrate the prediction workflow.

library(DiabetesPredictor)

# 1. Load the built-in sample data (first 10 rows of the dataset)
# Using system.file to locate the data within the installed package
data_path <- system.file("extdata", "sample.csv", package = "DiabetesPredictor")
input_data <- read.csv(data_path)

# Display raw input
head(input_data, 3)

# 2. Run Prediction
# The function automatically sanitizes the input and loads the internal model
predictions <- predict_diabetes(input_data)

# 3. View Results
print(predictions)

Model Information

• Algorithm: Random Forest
• Training Data: CDC Behavioral Risk Factor Surveillance System (BRFSS) 2015 https://www.kaggle.com/datasets/prosperchuks/health-dataset/data?select=diabetes_data.csv.
• Input Features: HighChol, HighBP, DiffWalk, Age, GenHlth, BMI, HvyAlcoholConsump, CholCheck.

Model Performance

The model's classification performance was evaluated using ROC (Receiver Operating Characteristic) and PRC (Precision-Recall) curves on the test set.

(Figure: ROC curve achieving an AUC of 0.82, indicating strong discriminative ability. The PRC curve (AUC=0.79) further validates performance on the imbalanced dataset.)

Key Metrics

• AUC (Area Under Curve): 0.8196
• Optimal Cutoff: 0.24 (Optimized for Recall)
• Sensitivity (Recall): 87.6%

Acknowledgements

• Xiaoyu Zhang: Model Trainer, R Coding
• Zhicheng Yang: Package Encapsulator, R coding