Smart Health: T2D Early-Stage Detection using ML

Objective

This project focuses on evaluating and comparing the performance of various baseline supervised machine learning models for early-stage detection of Type 2 Diabetes (T2D).

Dataset and Preprocessing

• Source: UCI Early Stage Diabetes Risk Prediction Dataset
• Records: 520
• Target: class (Positive/Negative)
• Preprocessing Steps:
  • Encoded features (Yes/No, Male/Female)
  • Scaled numeric features using StandardScaler
  • Train-test split: 80%–20%

Baseline Models Trained

• Logistic Regression
• Decision Tree
• Random Forest
• K-Nearest Neighbors (KNN)
• Support Vector Machine (SVM)
• Naive Bayes

Evaluation Summary

Model	Accuracy	Precision	Recall	F1 Score
Logistic Regression	0.942308	0.983333	0.921875	0.951613
Decision Tree	0.990385	1.000000	0.984375	0.992126
Random Forest	0.990385	1.000000	0.984375	0.992126
KNN	0.932692	0.983051	0.906250	0.943089
SVM	0.990385	0.984615	1.000000	0.992248
Naive Bayes	0.942308	0.967742	0.937500	0.952381

Visualizations

Accuracy Comparison

F1 Score Comparison

Confusion Matrix (All Models)

ROC Curve

Feature Correlation

Key Highlights

• Dataset cleaned and preprocessed with EDA notebook
• Models trained using scikit-learn
• Performance compared using accuracy, precision, recall, F1
• Visuals generated using matplotlib and seaborn

Author

Elizabeth Dada

GitHub: @edada2018
LinkedIn: linkedin.com/in/edada2018

References

The following studies contributed to the development and benchmarking of models in this project:

1. M. A. Bülbül, “A Novel Hybrid Deep Learning Model for Early Stage Diabetes Risk Prediction,” IEEE Access, vol. 12, pp. 45321–45333, 2024.
2. N. Fatima, S. A. Masud, and S. Muhammad, “Hybrid deep learning model for diabetes mellitus prediction,” IEEE Access, vol. 10, pp. 112233–112244, 2022.
3. P. Kaur, G. Kumar, and M. Kumar, “A healthcare monitoring system for diabetes prediction,” IEEE Rev. Biomed. Eng., vol. 14, pp. 290–299, 2021.
   → See full reference list