Kawasaki Disease (KD) is a rare heart condition that affects children all over the world. We use Kawasaki Disease patient data to train various machine learning models to predict whether a given patient has Kawasaki Disease or is a febrile control (i.e. does not have the disease). More information on Kawasaki Disease can be found here.
git clonethis repository;cdto repository directory- Create a symlink from KD-data Dropbox folder to
deep-learning-kd-diagnosis/data(this is confidential patient data not available to the public) - Create conda environment:
conda create -n kd;source activate kd - Install requirements:
pip install -r requirements.txt - Run experiments:
bash run_run_all.sh
- K-Nearest Neighbors (K-NN)
- Logistic Regression
- Support Vector Machine (SVM)
- Tree-Based Methods: Random Forest, XGBoost
- Deep Neural Network
- Ensemble (Voting/Bagging) Classifiers
- 5-Fold (Nested) Cross Validation for model selection and evaluation
- Metrics: Sensitivity, Specificity; ROC-AUC
- Python 3.x
- Numpy/Scipy
- Scikit-Learn
- Matplotlib
- Tensorflow
- XGBoost
- Fancyimpute
- Scikit-Optimize
- Imbalanced-Learn