Machine learning study from health record to remission

Impact of Hospital Health Record on the Remission of Diabetes Inpatients

README

Goal of this project：

According to the data of the hospital, try to identify a specific group of inpatients whose health improves. Original author found that HbA1c determination may improve patient outcomes and lower cost of inpatient care. We want to assess the importance of other factors (out of a list of 50 features) and compare the predictive performance of algorithms such as Decision Tree, Neural Networks and SVM.

Our project relies mainly on scikit-learn, Pandas, matplotlib, NumPy, HyperOpt, SciPy and seaborn.

The code contains the following files:

• ./data/diabetic_data_initial.csv <-- rawdata
• ./data/id_mapping.csv <-- raw id mapping
• ./results/*.html <-- output results of several runs
• ./preprocess.ipynb <-- notebook to preprocess data
• ./train_rf.ipynb <-- notebook to train random forest
• ./train_mlp.ipynb <-- notebook to train neural network
• ./train_svm.ipynb <-- notebook to train SVM

Notice

• The code contained in the jupyter notebook preprocess.ipynb should be run first.
• GPU is not required.
• Training takes 0.5~12 hours.
• The report notebook saves files to the "data" directory.

Reference

The diabetic_data_initial data was downloaded from https://www.hindawi.com/journals/bmri/2014/781670/#supplementary-materials