Optimasi Akurasi Koefisien Pajak Kendaraan Bermotor di Indonesia Menggunakan Metode Klasifikasi dan Regresi

Abstract

The growing awareness of the impact of motor vehicle emissions on the environment has encouraged Indonesia’s Ministry of Environment and Forestry to enforce emission testing regulations. These emission standards serve as a basis for calculating Motor Vehicle Tax (PKB). The Transportation Technology Research Center (BRIN) developed a tax coefficient prediction system to support this policy. Initial research utilized Orange Data Mining for machine learning analysis with algorithms like Random Forest, Neural Network, and AdaBoost. However, Orange Data Mining has limitations in flexibility, particularly in parameter tuning and preprocessing data, as well as inefficiencies in handling large datasets. This study adopts a more flexible approach, employing AutoML LazyPredict for quick identification of optimal models and GridSearchCV for hyperparameter optimization. The methodology involves two approaches: classification and regression. Classification employs models such as Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Tree, Extra Tree, and LightGBM, while regression utilizes Support Vector Regressor (SVR) optimized with GridSearchCV. Both approaches enable a comprehensive comparison and analysis of model performance. The results indicate that SVM and Decision Tree excelled in classification, achieving an accuracy of 100%. In regression, the models demonstrated high 16 performance with R² values ranging from 0.95 to 1.00, indicating exceptional predictive accuracy. Evaluations were conducted using metrics such as MAE, MSE, and R² for regression, along with accuracy scores and classification reports for classification tasks. This research underscores the effectiveness of machine learning model optimization, with both analyzed algorithms showing outstanding performance for classification and regression tasks.