Predicting Stroke from Electronic Health Records

With the spirit of reproducible research, this repository contains all the codes required to produce the results in the manuscript:

Soumyabrata Dev, Hewei Wang, Chidozie Shamrock Nwosu, Nishtha Jain, Bharadwaj Veeravalli, and Deepu John, A predictive analytics approach for stroke prediction using machine learning and neural networks, Healthcare Analytics, 2022.

Please cite the above paper if you intend to use whole/part of the code. This code is only for academic and research purposes.

Code Organization

All codes are written in R and python.

Code

The script to reproduce all the figures, tables in the paper are as follows:

• main.R: main scripts in R
• boxplots_adding.py: script to plot the boxplot of adding features
• boxplots_removing.py: script to plot the boxplot of removing features
• PCA8-evaluation.R: pca analysis stuffs
• top3-features.R: Results obtained by considering top-3 features
• More details.ipynb: Other experiments
• CHADS2_stroke_proportion.ipynb: Results about the proportion of cases with a stroke event as predicted by CHADS₂ for each score level respectively
• CNN_all_Features.ipynb: Results obtained by CNN model considering all features
• MPL_all_Features.ipynb: Results obtained by MPL model considering all features
• SVM_all_Features.ipynb: Results obtained by SVM model considering all features
• LASSO_all_Features.ipynb: Results obtained by LASSO model considering all features
• ElasticNet_all_Features.ipynb: Results obtained by LASSO model considering all features

Results

We also share the results obtained in our random downsampling experiments. The results obtained with the various benchmarking approaches is found in my_results.csv.