Rancang Bangun Aplikasi Peramalan Jumlah Penumpang Menggunakan Long Short-Term Memory (LSTM)

Abstract

Listen

Translate article

Public services such as public transportation are closely related to user satisfaction. Busway DKI Jakarta is one of the first public transportation services in Southeast and South Asia. In order to maintain passenger satisfaction, the management of Busway continues to improve services such as adding buses and opening a new line. Opening a new lane or adding buses must necessarily be adjusted also with the increasing number of passengers. So to know the number of passengers in the future, it is necessary to forecast the number of passengers through existing historical data. The historical data used is time series data from January 2015 to January 2016. The method used in forecasting is Long Short-Term Memory (LSTM), one of the machine learning methods. The method is measured in accuracy using Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE). As a comparison of LSTM accuracy in forecasting, we also use the Exponential Smoothing method. Based on the results of forecasting, the most and least dominant method of producing RMSE and MAPE is the LSTM method. Only in corridor 3 LSTM can not provide RMSE and MAPE values below baseline values. While corridor 5 LSTM can give better results after the data transformation process by using exponential smoothing. However, overall the LSTM method provides the best accuracy based on the lowest average RMSE and MAPE values, namely RMSE of 2640.53 and MAPE of 9.14%.