Model Hybrid SARIMA-ELM Untuk Peramalan Jumlah Penumpang Kereta Api di Stasiun Bandung

Abstract

This study aims to determine the application of the Hybrid SARIMA-ELM model to train passenger data and determine the level of forecasting accuracy. Forecasting is based on data on the number of passengers disembarking at Bandung Station in the period before Covid-19 (August 2014-December 2018) and the period during Covid-19 (January 2020-December 2021). Forecasting begins with finding the best SARIMA model, then the residuals from SARIMA are modeled with ELM. Hybrid SARIMA-ELM model forecasting results are measured for accuracy using MAPE. The best SARIMA model is SARIMA (0,1,1)(0,1,0)¹² for the phase before Covid-19 and ARIMA (2,0,0) model for the phase during Covid-19. With a data ratio of 80:20, 3 input features, 1 hidden layer neuron, and 1 output in the ELM model, a MAPE value of 5.27% is obtained, which means that forecasting is highly accurate for the model before Covid-19 and a MAPE of 13.43%, which means good forecasting for the model during Covid-19.

.

Keywords: ELM, Forecasting, Number of Passengers, SARIMA

Abstrak

Penelitian ini bertujuan untuk mengetahui penerapan model Hybrid SARIMA-ELM pada data jumlah penumpang kereta api serta mengetahui tingkat akurasi peramalannya. Peramalan didasarkan pada data jumlah penumpang yang turun di Stasiun Bandung pada periode sebelum Covid-19 (Agustus 2014-Desember 2018) dan periode saat Covid-19 (Januari 2020-Desember 2021). Peramalan diawali dengan mencari model SARIMA terbaik, kemudian residual dari SARIMA dimodelkan dengan ELM. Hasil peramalan model Hybrid SARIMA-ELM diukur keakuratannya menggunakan MAPE. Model SARIMA terbaik, yaitu SARIMA untuk fase sebelum Covid-19 dan model ARIMA (2,0,0) untuk fase saat Covid-19. Dengan perbandingan data 80:20, 3 fitur input, 1 neuron hidden layer, dan 1 output pada model ELM, diperoleh nilai MAPE sebesar 5,27% yang berarti peramalan berakurasi tinggi untuk model sebelum Covid-19 dan MAPE sebesar 13,43% yang berarti peramalan baik untuk model saat Covid