Machine learning (ML), also called self-learning or statistical learning, has become a method of choice for many research projects concerned with predictions and pattern recognition in computer science, natural sciences, economics and medicine. It has also been a transformative technology for data-driven enterprises like Facebook, Twitter or Google.
Social sciences, political sciences and humanities on the other hand remain almost entirely excluded from this development with very few papers and contributions even considering or proposing machine learning techniques for current research questions. Maybe the presumed conclusion that machine learning is only viable for large-scale big data projects is so prevalent that social scientists remain fixated on other forms of comparative analysis for quantitative data?
The proposed paper tests these new methods and applies ML techniques to macro-level conflict research with the objective of correctly predicting the conflict intensity of all countries 12 months into the future based on a comparatively small-scale dataset of seven available socio-economic and political variables.
By accumulating data from various years to a set of 851 cases and controlling these individual country-years for autocorrelation, this paper intends to demonstrate the relative importance of the socioeconomic framework of states for the outbreak of violence as well as the tremendous but currently untapped potential of ML and Deep Learning Techniques for Peace and Conflict Research.
By building a classification framework which can estimate the correct conflict intensity with unanticipated success rates of up to 80% for new cases, this approach achieves a predictive quality outperforming any regression-based predictive technique by a wide margin. Using Random Forest, a classification tree method and k-Nearest Neighbor Methods an instance-based method, two very differently operating ML algorithms are tested for their advantages and potential application in quantitative research projects of social and political sciences.
Furthermore, the proposed framework could be used as a foundation for real-time predictions of upcoming socio-economic conflict risks in future years if the process of variable gathering can be accelerated or predictions of variables can be used in modelling different development scenarios.
