Machine Learning Classification of Peaceful Countries: A Comparative Analysis and Dataset Optimization

We used a machine learning approach to classify countries as peaceful or non-peaceful by analyzing linguistic patterns in global media articles. Utilizing vector embeddings and cosine similarity, we develop a supervised classification model that achieves 94% accuracy in distinguishing peaceful contexts. We also examine the impact of dataset size on model performance, finding that while larger datasets improve accuracy, smaller datasets still provide valuable insights. Our calculated peace percentages correlate strongly with the Human Development Index (HDI), validating our methodology. This study highlights the potential of advanced AI techniques in peace studies and underscores the importance of data quality and representativeness. Future work includes developing real-time monitoring tools and addressing training data biases to enhance model accuracy.