The Use of Machine Learning Methods in Political Science: An In-Depth Literature Review.

This article provides a systematic review of how machine learning methods are currently applied in political science research, analyzing 339 peer-reviewed articles published between 1990 and 2022. It finds a significant increase in the use of machine learning over the past decade, particularly in political communication, conflict and peace studies, and policy studies, with text-as-data being the predominant data type analyzed. The most common machine learning methods include topic modeling (an unsupervised learning technique), random forest, and support vector machines (both supervised learning methods), while reinforcement learning is notably absent. Although machine learning is primarily used for predictive purposes, explanatory applications and model explainability remain limited, and only a minority of studies report best practices such as hyperparameter tuning or benchmarking. The review also highlights that most research is authored by political or social scientists rather than interdisciplinary teams, and it calls for greater collaboration and methodological rigor to enhance reproducibility and the effective use of advanced machine learning techniques in political science.