Rhee, Jin-Woo, and Han-Jin Lee. "A Study on the Classification Criteria for Automotive Electronics Using Random Forest." Digital Trade Review 23, no.1 (2023): 31-52.
-Abstact-
This study aims to propose appropriate classification criteria for automotive electrical and electronic components (automotive electronics). Using Random Forest, a machine learning model, this study analyzed classification case data from the Korea Customs Service. The HS code at the 2-digit (chapter) and 4-digit (heading) levels in this data were set as the target variables, while the description, reasoning, and conclusion of each classification case were used as the independent variables. Six models were constructed by combining different target and independent variables. The hyperparameters of each model were optimized using GridSearchCV, and their performance was evaluated based on accuracy, precision, recall, and overfitting. As a result, the HS4-DRC and HS2-DRC models were selected, and the Random Forest analysis results were presented. This study conducted a feature importance analysis to identify the key factors contributing to the classification of automotive electronics. The major contributing features included ‘measurement’, ‘flow rate’, ‘inspection’, ‘control’, ‘circuit’, ‘detection’, ‘temperature’, ‘recording’, and ‘resistance’. Based on this analysis, classification criteria were proposed, focusing on the functional and technical characteristics of automotive electronics as well as their intended use. Given the complexity and diversity of automotive electronics, consistent and legally compliant classification poses challenges. This study suggests a classification framework using Random Forest and aims to ensure that classification does not become a trade barrier.