Forecasting Demand for Motorbikes at Astra Motor Balikpapan Using Support Vector Regressor

Abstract

Forecasting requests for motorbikes is a critical aspect of Astra Motor Balikpapan's operations. The Support Vector Regression (SVR) model, a method commonly used in forecasting, is particularly useful when dealing with complex data that may contain outliers and when the data is limited. This research evaluates the performance of the SVR model in estimating requested motorbikes at Astra Motor Balikpapan for 3, 6, 9, and 12 months, and analyzes the impact of parameter changes in the model evaluation. The request data for Astra Motor Balikpapan motorbikes used for five years or 60 months, which are divided into two parts: training and test data. The SVR model was built with three Kernel types: linear, polynomial, and RBF kernels. The evaluation results demonstrate the SVR model's ability to predict request motorbike with Sufficient accuracy, with minor mark errors, including an average MAE of about 0.49, RMSE of about 0.58, and R² score of about 0.99. Parameter changes also affect model evaluation, as in the case of ADV motorbike with RBF kernel; adjustment of parameter C from 0.01 to 10 results in significant accuracy, decreasing MAE from 0.36 to 0.004. This study concludes that the SVR model is an effective method for predicting motorcycle requests, with practical implications for Astra Motor Balikpapan's operations.