merelwieberdink/thesis_code

Binary classification on survey data using ML methods: Predicting perception of police corruption African Countries # thesis_code Binary classification on survey data using ML methods: Predicting perception of police corruption African Countries # Description The project is the implementation of a thesis of the master in Data Science and Society at Tilburg University. This project implements machine learning methods to predict perception of police corruption in African countries. It uses survey data from the Afrobarometer Merged Round 8 Data (34 Countries), which is publicly available on their website: afrobarometer.org. The project includes the comparison of the predictive models logistic regression, random forest, XGBoost, and muli-layer perceptron. It also includes feature importances, error analysis, and an evaluation of the different countries in the dataset and potential bias (including separate country performance and class distributions). The python notebook file is divided into three main parts, with several (layers of) subsections. These three main sections are: (1) Pre-processing (which has been partly performed in R prior, and not included here), (2) modeling, and (3) out-of-sample evaluation and error analysis (including the country comparison).