Machine Learning Approaches for Export Trend Classification: Evidence from Leading Commodities in Indonesia

Abstract

Sorong City holds a strategic position in the export economy of Papua Barat Daya; however, its export performance remains volatile due to global price fluctuations, logistical constraints, and shifts in international demand. To address these challenges, this study applies machine learning-based classification to analyze and predict export trend dynamics of Sorong’s leading commodities. Specifically, the study compares the performance of Naïve Bayes and Random Forest classifiers within a quantitative experimental framework. The dataset comprises 874 export records (2023–2025), including HS Codes, export values, destination countries, exporters, and export types. The methodological workflow encompasses data preprocessing, trend labeling, normalization, label encoding, class balancing using SMOTE, and model evaluation via 80:20 train-test split and 10-fold cross-validation. Performance metrics include accuracy, precision, recall, F1-score, and ROC-AUC. Experimental results reveal that Random Forest outperforms Naïve Bayes, achieving 74% accuracy compared to 57%, and more effectively captures nonlinear feature relationships. Despite a reduction in ROC-AUC during cross-validation, Random Forest demonstrates greater robustness in export trend prediction. Overall, the findings highlight the potential of machine learning to enhance regional trade forecasting, inform evidence-based policy formulation, and strengthen data-driven export management in emerging regional economies.