Label Ranking for Election Outcome Prediction (MSc)
Predicting the election outcome is a complex task. Currently used approaches, such as classic pools, might not be able to capture the multitude of variables involved in the election results. This work proposes tackling the problem of Election Outcome Prediction with the use of Data Mining and Machine Learning. The focus will be in the study of the preferences of the parties and their socio-economic context. To tackle this, we suggest to transform the prediction of elections into a Label Ranking (LR) problem.
LR is a subtask of the Preference Learning field, which uses a set of tools that allow the discovery of patterns in preferences. This has the advantage of allowing the prediction of, not only who is going to win, but also an ordered relation between the political parties or candidates. In particular, we are going to focus on Pairwise Association Rules (PAR). We will use them for prediction purposes. They come with the advantage that they provide interpretable results, which is useful to analyze the predictions.
The results will be tested both in common LR datasets and in election datasets. We will compare our approach with other LR algorithms. In the end, considering the good results obtained, we believe that this work holds promise both as a contribution to the LR community and the Political Science field.
run
R /src/core.R