<![CDATA[The escalation of conflicts in the Asia-Pacific region risks triggering regional destabilization, yet evidence-based approaches to conflict prevention remain limited. This study aims to develop a machine learning framework to analyze the level of conflict escalation and support peacebuilding. Using a quantitative comparative design, secondary ACLED data from 2010–2025 were analyzed through a workflow encompassing data preprocessing, feature engineering, handling class imbalance using SMOTE, and evaluating four algorithms: Random Forest, XGBoost, Linear SVM, and Multi-layer Perceptron. Model performance was measured using accuracy, precision, recall, weighted F1 score, and macro AUC. The results showed that Random Forest performed best (accuracy 91.4%; weighted F1 score 91.5%; macro AUC 0.94), followed by XGBoost (accuracy 89.9%; macro AUC 0.93). Both ensemble models excelled in distinguishing low and moderate escalation classes, but their performance declined in the high and critical categories due to limitations in the original sample. Feature importance analysis revealed event type and frequency as the most significant predictors. These findings confirm the effectiveness of tree-based methods for structured conflict classification and their contribution to the development of early warning systems for conflict prevention in the Asia-Pacific region. The proposed framework supports anticipatory decision-making, timely intervention, and more efficient resource allocation for policymakers and international organizations.]]>
