Machine driven predictions of the socio-economic status of Twitter users

In this study, we examine the possibilities of creating predictions of the Socio-Economic Status of twitter users with the use of machine learning algorithms. The study starts with an ethical evaluation of the process and data collection. After the data is collected we explore the data and create features and labels. We then employ machine learning algorithms out of the box and combine the prediction probabilities via a soft-voting ensemble method. We find that the model outperforms dummies, but has troubles with the skewed data. We run additional tests to examine the performance of the classifiers when the tasks increase in probability, where we find the random forest to show the most promising results.