Chronic-Kidney-Disease-Prediction

The goal of this project is predit the chronic kidney disease using parameters such as 'sugar', 'bacteria', 'blood_glucose_random', 'blood_urea',, 'white_blood_cell_count', 'hypertension', 'diabetes_mellitus', 'coronary_artery_disease', etc.

Description

Domain : Healthcare
Technology:- EDA, Feature Engineering, Feature Selection and Machine Learning
Languages:- python

Libraries used:-
Pandas Numpy Matplotlib Seaborn Plotly Sklearn

1. Dataset:- Kidney_data.csv

This dataset can be used to predict the chronic kidney disease and it can be collected from the hospital nearly 2 months of period.

Attributes in given data set:-

age - age
bp - blood pressure
sg - specific gravity
al - albumin
su - sugar
rbc - red blood cells
pc - pus cell
pcc - pus cell clumps
ba - bacteria
bgr - blood glucose random
bu - blood urea
sc - serum creatinine
sod - sodium
pot - potassium
hemo - hemoglobin
pcv - packed cell volume
wc - white blood cell count
rc - red blood cell count
htn - hypertension
dm - diabetes mellitus
cad - coronary artery disease
appet - appetite
pe - pedal edema
ane - anemia
class - class

2. Data Cleaning

Steps followed during data cleaning:-

1. Replace all the column names with their full name for better understanding.
2. Replace categorical values into numerical values.
3. Correct the mis-spelled categorical values.
4. Replaced the null values with median for numerical columns and mode for categorical columns.
5. Removing Duplicated rows.

3. Exploratory Data Analysis and Data Pre-Processing

Steps followed in EDA:-

1. Performed Univariate Analysis, Bivariate and Multivariate Analysis to Understand the data and find some patterns.
2. Prepared data for model training by converting all categorical columns to numerical using LabelEncoder.
3. Feature Selection using Variance Inflation Factor (VIF).
4. DIVIDE THE DATA INTO X (Independent variable), y (target variable / dependent variable).
5. Spilt the 80 % data for training and 20% data for testing.

5. Creation of machine learning model

Machine Learning models are created by following Algorithms

1. Logistic Regression
2. Decision Tree Classifier
3. SVM Classifier
4. Gaussian Naive Bayes
5. Random Forest Classifier
6. Ada Boost Classifier
7. Gradient Boosting Classifier
8. XG Boost Classifier

6. Hyperparameter tunning

Perform the hyperparameter tuning to avoid the overfitting in model.

7. Model Evaluation

We got 97.5% accuracy score in SVM, XGBoost Classifier, Ada Boost Classifier, and Random Forest Classifier

Insights

EDA and Feature Engineering:

1. Demographics and Symptoms:

   • CKD predominantly affects older individuals, with blood pressure increasing significantly with age.
   • Symptoms like peda edema, poor appetite (in 20.6% of cases), and anemia are common among CKD patients.

2. Co-Morbidities:

   • A strong correlation exists between diabetes mellitus, hypertension, and CKD, highlighting their combined influence on kidney health.
   • 34.4% of the dataset shows individuals with diabetes, and a significant number of these also exhibit hypertension.

3. Biomarkers:

   • Serum creatinine and blood urea levels are markedly higher in CKD-positive cases, with significant outliers indicating varying stages of the disease.
   • Elevated albumin levels in individuals with diabetes suggest kidney damage as a direct result of the condition.

4. Feature Selection:

   • Age, blood pressure, albumin, sugar, serum creatinine, and diabetes mellitus were identified as the most impactful features for CKD prediction.

Model Training and Evaluation:

1. Model Performance:

   • SVM Classifier and XGBoost Classifier achieved the highest accuracy (97.5%), effectively handling the complex relationships in the dataset.
   • Ensemble models like Random Forest and AdaBoost delivered 97.5% accuracy, indicating strong predictive capabilities.

2. Impact of Hyperparameter Tuning:

   • Fine-tuning parameters significantly improved accuracy across models:
     • SVM Classifier: Improved from 91.25% to 97.5%
     • Logistic Regression: Improved from 90% to 96.25%.
     • Gradient Boosting and AdaBoost showed enhanced generalization after tuning.

3. Evaluation Metrics:

   • Models consistently showed balanced precision, recall, and F1-scores, with minimal false negatives, essential for early CKD diagnosis.

4. Overfitting:

   • Ensemble models showed slight overfitting, as evidenced by near-perfect training accuracy compared to test accuracy, requiring further validation on external datasets.

Summary

• Key Insights:

  • Diabetes mellitus, hypertension, and elevated serum creatinine are critical factors influencing CKD.
  • Ensemble models, especially XGBoost, excel in predictive accuracy and reliability for CKD detection.
  • Age-specific trends (e.g., rising blood pressure with age) and symptoms like anemia and peda edema provide valuable clinical insights.

• Workflow:

  • The project followed a structured pipeline:
    1. Data cleaning
    2. Comprehensive exploratory analysis.
    3. feature engineering and Preprocessing.
    4. Model training and hyperparameter tuning.

Conclusion

1. Critical Drivers:

   • Biomarkers like serum creatinine, albumin, and blood urea are key predictors of CKD, with age, hypertension, and diabetes serving as significant contributing factors.
   • Co-morbidities should be monitored closely for early intervention.

2. Model Deployment:

   • XGBoost, with its superior accuracy, is recommended for real-world CKD prediction systems.
   • SVM Classifier, Random Forest and AdaBoost can serve as robust alternatives, offering high performance and interpretability.

3. Future Directions:

   • External validation on larger datasets to improve model generalization.
   • Development of lightweight models for deployment in resource-limited settings.
   • Collaboration with medical professionals to enhance model explainability and utility in clinical workflows.

📩 Feedback

If you have any feedback, please reach out to me at Linkedin: https://www.linkedin.com/in/vineet-patel416/