Predictive Performance Evaluation of ARIMA and Hybrid ARIMA-LSTM Models for Particulate Matter Concentration
This study provides an objective evaluation of prediction performance models for particulate matter policy for industrial stakeholders by comparing the ARIMA and Hybrid ARIMA-LSTM models for predicting air quality data from the industrial environment. In the case of PM 1.0 concentration, we have an...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Department of Informatics, UIN Sunan Gunung Djati Bandung
2024-12-01
|
| Series: | JOIN: Jurnal Online Informatika |
| Subjects: | |
| Online Access: | https://join.if.uinsgd.ac.id/index.php/join/article/view/1318 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This study provides an objective evaluation of prediction performance models for particulate matter policy for industrial stakeholders by comparing the ARIMA and Hybrid ARIMA-LSTM models for predicting air quality data from the industrial environment. In the case of PM 1.0 concentration, we have an RMSE value of 8.29 and an error ratio of 0.45 for the ARIMA model and an RMSE value of 3.54 and an error ratio of 0.22 for the hybrid ARIMA-LSTM model. Meanwhile, for PM 2.5 concentration, we obtain an RMSE value of 6.61, an error ratio of 0.66 for the ARIMA model, an RMSE value of 2.68, and an error ratio of 0.19 for the hybrid ARIMA-LSTM model. According to this study, the ARIMA model, which is found in autoarima and represents the best model, is (2,0,1) for PM1.0 and (1,0,1) for PM2.5. The hybrid ARIMA-LSTM model outperforms the ARIMA model in terms of prediction accuracy, as evidenced by the RMSE and error ratio values, which are improved by approximately 57.30% and 51.11% for PM1.0 and 59.46% and 71.21% for PM2.5, respectively, since the hybrid ARIMA-LSTM model can accommodate variable-length sequences and capture long-term relationships to become noise-resistant, which makes higher prediction accuracy possible. |
|---|---|
| ISSN: | 2528-1682 2527-9165 |