Knowing our choices: unveiling true voting patterns through machine learning (ML) and natural language processing (NLP) in European Parliament

Abstract Anticipating the diverse voting behavior of European Parliament Members (MEPs) is complex due to their diverse backgrounds, affiliations, countries, and ideological stances. These factors result in varied preferences, posing a substantial analytical challenge in understanding and identifying significant patterns. This research is the pioneering venture in this field, specifically within the European Parliament (EP) context. To bridge this research gap, the study aims to forecast MEPs’ voting behavior using semantic descriptors of legislative proposals through Machine Learning (ML), contributing insights to both Machine Learning and political science domains. Random Forests, Support Vector Machines (SVM), and Logistic Regression models are used and compared against a baseline to evaluate their predictive performance, where Random Forests emerged as a top model achieving 73–74% accuracy. The study also examines how MEPs’ affiliations and ideologies impact voting predictability. Notably, extreme political ideologies correlate negatively with accuracy. Analyzing words and topics in legislative proposals, findings align with the shift after the 2014 $$8\text{th}$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:mn>8</mml:mn> <mml:mtext>th</mml:mtext> </mml:mrow> </mml:math> EP election. The primary focus transitioned from left-right ideologies to pro and anti-European integration stances. Finally, the study highlights the potential of Machine Learning and Natural Language Processing (NLP) techniques combined with semantic descriptors of the laws to predict MEPs’ voting behavior. Graphical abstract