Skip to content

Using machine learning models to predict if patients have chronic kidney disease based on a few features. The results of the models are also interpreted to make it more understandable to health practitioners.

License

Notifications You must be signed in to change notification settings

Shuyib/chronic-kidney-disease-kaggle

Repository files navigation

chronic-kidney-disease-kaggle

The data consists of 25 features and 400 observations consisting of various features that could be used to predict chronic kidney disease. The original dataset is available UCI Machine Learning repository and you can find the kaggle dataset here. I use data cleaning and data transformation techniques in cluding data compression techniques for example PCA (Principal Component Analysis) to evaluate the performance of Logistic regression and decision tree classifiers.

Here are the final scores:

Logistic regression without standard scaler: F1-Score: 0.992
Decision tree classifier without standard scaler: F1-Score: 0.959

Logistic regression with standard scaler: F1-Score: 0.976
Logistic regression with standard scaler and PCA: F1-Score: 0.625
Decision tree classifier with standard scaler and PCA: F1-Score: 0.879

Within the notebook we go over feature importance for each classifier and also interpret what the results mean to the patient and the progression of the disease. How they relate to diagnostic tests for kidney function expose about the patients' condition? Just for reading use this. You can also make my kaggle environment using the pip install -r requirements.txt. Use a virtual environment so as not disturb what you already have learn how to create one using the official documentation or using anaconda.

Lastly they are tasks you can try out to further improve and refine the models further. Check them out maybe you'll learn something. Feel free to reach out to talk about the results further.

Other things you can try:

  • Fix class imbalance with Synthetic Minority Oversampling Technique SMOTE.
  • Apply decision boundaries better test.
  • Check out the best practices applied in this examples
  • This new classifier looks promising LinearBoostClassifier

How to run the docker image

Build docker image

docker build --platform linux/amd64 -t chronic-kidney-disease-kaggle .

Run the Docker image

docker run -it --platform linux/amd64 -p 9999:9999 chronic-kidney-disease-kaggle:latest