Prediksi Emisi Co2 Di Indonesia Menggunakan Pendekatan Hybrid Arima Dan LSTM

Abstract

Climate change has emerged as a pressing global issue, with carbon dioxide (CO2) 
emissions serving as a major contributor to global warming. In Indonesia, the expansion of industrial 
activities, transportation, and the reliance on fossil fuel-based energy have significantly accelerated 
CO2 emission levels. In this context, the need for accurate emission forecasting has become 
increasingly important as a basis for formulating data-driven mitigation policies. This study aims to 
develop a predictive model for CO2 emissions in Indonesia using a hybrid approach that combines 
AutoRegressive Integrated Moving Average (ARIMA) and Long Short-Term Memory (LSTM) methods. 
ARIMA is employed to capture linear patterns in historical time series data, while LSTM is used to model the non-linear and complex dynamics often present in environmental data. The emission data 
used spans from 1970 to 2023, with training and testing data separated chronologically in an 80:20 
ratio. The evaluation results show that the ARIMA model alone yielded suboptimal performance 
(RMSE: 2342.5139, MAE: 2341.5775, MAPE: 414.77%), whereas the LSTM model significantly 
improved prediction accuracy (RMSE: 49.3307, MAE: 45.5498, MAPE: 7.94%). The hybrid ARIMA
LSTM model achieved the best results, with an RMSE of 31.5778, MAE of 25.0335, and MAPE of 
4.34%. These findings indicate that the combination of both methods substantially enhances prediction 
performance compared to standalone models. The implications of this research are twofold: 
academically, it contributes to methodological development in environmental data analysis; 
practically, it offers valuable insights for policymakers in formulating more effective and sustainable 
carbon emission reduction strategies in Indonesia.