DETEKSI POLA SKRAP PLASTIK DALAM DATA PERDAGANGAN GLOBAL

MUHAMMAD SUKRI BIN RAMLI (ROYAL MALAYSIAN CUSTOMS DEPARTMENT)

JURNAL PERSPEKTIF BEA DAN CUKAI
Vol. 9 No. 3 (2025): Regional Training Center

Cheap, contaminated plastic scraps are often mislabelled under high-value HS codes, skewing global trade data and weakening agreements like the Basel Convention. We detect this fraud by identifying an “inverse price -volume signature” a pattern where reported volumes climb as unit prices fall. Our transparent machine-learning pipeline analyses UN Comtrade HS?39 data (2020-2024), engineers both basic and advanced price -volume measures, groups products with K-Means clustering, and applies a Random Forest model that flags high-risk segments with 93.75 % accuracy (0.89 precision, 0.92 recall). Explainable AI shows that price drops, volatile pricing, and rising volumes drive these alerts. Testing on firm-level records (2019-2025) confirms that global red flags especially for HS 390210 translate into actionable watchlists. This scalable framework equips regulators with a risk-based inspection tool under policies like Malaysia’s 2025 HS 39.15 Certificate of Approval and offers data-driven support for the international Plastics Treaty.