- Public dataset from the Cline Center at the University of Illinois, listing details of all known coups d'etat from the 1940s through 2004
- Data is almost entirely "one-hot encoded" as binary values (1 or 0) describing categorical variables
- New variables were constructed, e.g., decade, season
- A second dataset consisting of country-by-country global economic history data was merged with coup data
- Exploratory data analysis with correlations and visualizations
- Two R scripts were created using only the Cline Center coup dataset to predict the following outcomes:
- The success or failure of attempted coups
- The violence or lack thereof of attempted coups
- Another R script was created to predict coup events based on country-by-country economic indicators over time
- Numerous machine learning algorithms were used to model these outcomes
- Logistic regression
- Stepwise forward selection of model features for logistic regression
- Naive Bayes classification
- Decision trees classification
- Random Forest ensemble learning
- Lastly, out-of-sample data on recent coups provides another, mostly successful, test of the models
- The correlation matrix for coup success is visualized as follows:
