Investigation of graph-based clustering approaches along with graph neural networks for modeling armed conflict in Bangladesh

Determining fatality rates—a critical component of conflict analysis and comprehending the dynamics of armed conflict in Bangladesh are the main goals of this study. The contributions of this paper are twofold. Firstly, this paper presents the performance of graph neural networks (GNNs) is measured against traditional machine learning (ML) algorithms, specifically artificial neural networks (ANNs), for fatality prediction. Secondly, it has presented an empirical analysis of graph-based clustering and traditional clustering methods for modeling armed conflicts in Bangladesh. Predictive fatality models have been developed using graph neural networks and traditional machine learning algorithms. Graph neural networks exploit the graph structure of the data, allowing for more effective modeling and prediction. Traditional machine learning algorithms, and artificial neural networks, serve as benchmarks for comparison. The results indicate that graph-based clustering methods provide valuable insights into the structure and dynamics of armed conflicts in Bangladesh. The comparative analysis demonstrates that graph neural networks outperform a few traditional machine learning algorithms in terms of predictive fatality, highlighting the importance of capturing the relational dependencies present in the data. Graph-based clustering techniques offer an innovative approach by capturing the inherent relational structure of conflicting data, whereas traditional clustering methods rely on traditional feature engineering and distance metrics. A comprehensive dataset from ACLED for armed conflict incidents in Bangladesh was used to evaluate the performance. Both graph-based and traditional clustering methods were applied to identify meaningful clusters and analyze the patterns of armed conflict. An additional contribution has been made through the construction and analysis of a knowledge graph, elucidating the interconnected relationships among various datasets. In conclusion, this research contributes to the field of conflict analysis by showcasing the advantages of graph-based clustering methods and graph neural networks for understanding armed conflict patterns and predicting fatality rates.