Spaces:
Sleeping
A newer version of the Gradio SDK is available: 6.31.0
title: CKD Prediction System
emoji: 🫘
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 5.23.3
app_file: app.py
pinned: false
license: mit
Chronic Kidney Disease (CKD) Prediction System
Research Project | PAF-IAST | HEC-Funded
Principal Investigator: Dr. Muhammad Rizwan
Overview
This web application provides an AI-powered screening tool for Chronic Kidney Disease (CKD) based on routine blood test parameters. It was developed as part of a peer-reviewed research study conducted at the Pakistan Air Force Institute of Aerospace Technology (PAF-IAST) under an HEC-funded project.
The application accepts 21 blood parameters as input and returns a binary prediction (CKD / No CKD) along with class probabilities.
Model
| Property | Details |
|---|---|
| Algorithm | Random Forest Classifier |
| Accuracy | ~95.5% |
| F1 Score | ~95.4% |
| ROC-AUC | ~98.5% |
| Validation | 5-fold Stratified Cross-Validation |
| Explainability | SHAP + LIME |
The Random Forest model was selected after benchmarking 8 classifiers including XGBoost, CatBoost, Gradient Boosting, AdaBoost, Logistic Regression, Decision Tree, and K-Nearest Neighbors.
Input Features
The model uses the following 21 blood test parameters:
Demographic: Sex, Age
CBC: Haemoglobin, Hematocrit, RBC Count, WBC Count, Platelet Count
Red Cell Indices: MCH, MCHC, MCV, RDW-SD, MPV, PCT, PDW
Differential Count: Lymphocytes, Monocytes, Polymorphs, Eosinophils
Kidney Function: Creatinine, Blood Urea, GFR
Key Findings (from SHAP Analysis)
Top predictive features (in order of importance):
- Creatinine — strongest CKD indicator
- GFR (Glomerular Filtration Rate) — inverse relationship with CKD
- Urea — elevated in kidney dysfunction
- RDW-SD — red cell distribution width
- MPV — mean platelet volume
Dataset
- Samples: 398 patients
- Class distribution: CKD: 241 (60.6%) | No CKD: 157 (39.4%)
- Missing values: < 1.5% (imputed with zero-fill)
- No duplicates
Files in this Space
app.py ← Gradio application
requirements.txt ← Python dependencies
Random_Forest.pth ← Trained Random Forest model (joblib format)
README.md ← This file
Disclaimer
This application is intended for research and educational purposes only.
It does not constitute medical advice, diagnosis, or treatment.
All clinical decisions must be made by a licensed and qualified physician.
Citation
If you use this work, please cite the associated research paper (details to be added upon publication).
Contact: Dr. Muhammad Rizwan | PAF-IAST