Optimizing LSTM Model for Low-Cost Green Car Demand Forecasting

Abstract

Demand forecasting is an important task in every business including car manufacturing. The high initial production cost of cars places even more importance on demand forecasting especially for a specific type of car such as the Low-Cost Green Car (LCGC). Within its current 8 years journey, the number of demands for LCGC cars has experienced some fluctuation which makes the need for accurate demand forecasting even more important. This research aims to accurately predict the demand for LCGC cars in Indonesia using the Long Short-Term Memory (LSTM) method. However, it is difficult to find the best parameter settings for a neural network-based model such as LSTM. Therefore, this research will explore the effect of different parameter settings on the model accuracy. The data used in this research is the number of monthly domestic LCGC car sales from September 2013 to December 2021 obtained from the Association of Indonesian Automotive Industries (GAIKINDO). The experiments were conducted using the Tensorflow package in Python and were evaluated for their performance using MAE and MAPE. The experimental results showed that the LSTM model can accurately predict car sales/demands with an MAE of up to 977.6 and MAPE of 6.8% (accuracy 93.2%).