• Shiny app for predicting diabetes risk
• Based on health indicators from BRFSS 2015 (Kaggle)
• Development process:
  • Start with 21 health indicators
  • Used 20% stratified sample due to memory constraints
  • Applied XGBoost model for feature selection
  • Selected top 5 most impactful featuresUses XGBoost model
• Provides risk category and confidence score

# Load the data
data <- read.csv("diabetes_012_health_indicators_BRFSS2015.csv")
data <- data %>%
  select(c("HighBP", "BMI", "DiffWalk", "HighChol", "GenHlth", "Diabetes_012")) %>%
  mutate(across(!c(BMI), factor))
levels(data$Diabetes_012) <- c("Normal", "Prediabetes", "Diabetes")

(omit some codes)

# Display the summary table
kable(summary_data, digits = 2,
      col.names = c("Diabetes Status", "Count", "High BP", "High Cholesterol", "Avg BMI", "Avg Gen. Health"),
      caption = "Summary of Key Health Indicators by Diabetes Status (20% Sample)")

• User inputs:
  • BMI
  • High Blood Pressure
  • High Cholesterol
  • Difficulty Walking
  • General Health
• Output:
  • Risk category(Normal, Prediabetes, Diabetes)
  • Confidence score

# XGBoost Modelling
set.seed(8-26-2024)
final_xgb_model <- train(Diabetes_012 ~ .,data = sampled_data,method = "xgbTree")
saveRDS(final_xgb_model, "final_diabetes_xgb_model.rds")

# Load the model
model <- readRDS("final_diabetes_xgb_model.rds")

# Display feature importance
print(model$results[which.max(model$results$Accuracy), ])

##    eta max_depth gamma colsample_bytree min_child_weight subsample nrounds
## 18 0.3         1     0              0.8                1         1     150
##     Accuracy     Kappa  AccuracySD    KappaSD
## 18 0.8470395 0.1743532 0.002090137 0.01446783

## Create test set by randomly sampling 100 instances from the remaining data
remaining_data <- data[-train_index, ]
test_index <- sample(1:nrow(remaining_data), 100)
test_data <- remaining_data[test_index, ]

# Make predictions
predictions <- predict(model, newdata = test_data)

# Calculate performance metrics
confusion_matrix <- confusionMatrix(predictions, test_data$Diabetes_012)
cat("Test Accuracy:", confusion_matrix$overall['Accuracy'], "\n")

## Test Accuracy: 0.87

Remember: This app is for educational purposes only. Always consult healthcare professionals for medical advice.