Integrating XAI for Predictive Conflict Analytics

Predicting global conflicts through data-driven approaches has the potential to aid political decision- makers in formulating more effective and targeted policies. However, high-performance models that derive patterns from data often become highly complex, making it challenging to extract understandable rationales behind their outcomes. In this paper, we suggest integrating a transformer-based Artificial Intelligence Early Warning System (AI-EWS) with integrated gradients, an eXplainable Artificial Intel- ligence (XAI) technique attributing model predictions to specific features at a given time in the input data, thereby enhancing interpretability. To validate our methodology, we conduct experiments on a prominent geopolitical dataset: ACLED. This dataset provides comprehensive insights into global conflict events, facilitating effective pattern learning and generalization by our model. Leveraging these explainability techniques, our goal is to bridge the gap between complex, high-performance models and the practical needs of policymakers in conflict prevention and resolution. Predictive analytics algorithms in conjunction with an XAI approach can foresee the impact of decisions on various population segments, fostering equity, and inclusion and supporting a data-driven approach, along with a culture of openness and accountability within the public administration.