European parliament is the major political legislative body of the EU that causes divides
in public opinion since its beginning. While some of its opponents usually point out its
political and culutural heterogenity as a major weakness and cause of inability to function
eciently, others claim that MEPs often quickly lose their connection with voters and vote
mainly as it is directed to them by the leaders of their politival groups. We wanted to
nd out if MEPs voting patterns are predictable enough to be successfully predicted with
machine-learning based computer model.
To reduce time complexity of the problem we rather focused on joint votes of (national)
political parties than individual MEPs. At rst we implemented web crawlers that we
used to extract as many roll-call voting oriented data as we can. Than we combined data
mining with expert geopolitcal approach to extract the features and build a model for
voting prediction.
Our predictions were overall nearly 80% successful (weighted f1, roc-auc), however results
vary greatly between political groups. It became clear that we could easily predict votes
of coallition parties with liberal-globalist political orientation while eurosceptic, economic
social and nationalist parties seemed to be much more unpredictable. With that information
we further backed the importance of the new nationalist-globalist political cleavage. A challenge to the model presents also a class of votes of abstention, which is hard
to successfully predict even with expert human-knowledge.
|
