Machine Learning for Kawasaki Disease Diagnosis

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.

Usage Instructions:

1. git clone this repository; cd to repository directory
2. Create a symlink from KD-data Dropbox folder to deep-learning-kd-diagnosis/data (this is confidential patient data not available to the public)
3. Create conda environment: conda create -n kd; source activate kd
4. Install requirements: pip install -r requirements.txt
5. Run experiments: bash run_run_all.sh

Methods Evaluated:

• K-Nearest Neighbors (K-NN)
• Logistic Regression
• Support Vector Machine (SVM)
• Tree-Based Methods: Random Forest, XGBoost
• Deep Neural Network
• Ensemble (Voting/Bagging) Classifiers

Evaluation Methodologies/Metrics:

• 5-Fold (Nested) Cross Validation for model selection and evaluation
• Metrics: Sensitivity, Specificity; ROC-AUC

Dependencies:

• Python 3.x
• Numpy/Scipy
• Scikit-Learn
• Matplotlib
• Tensorflow
• XGBoost
• Fancyimpute
• Scikit-Optimize
• Imbalanced-Learn