Evidence record · UDINUS · SDGs
- Unit
- Rektorat UDINUS
- Period
- 2026
- Goals
- Record
- e756c3a9-2257-44a9-8854-031f28781a37
- Published
- 2026-10-30
Universitas Dian Nuswantoro (UDINUS) conducted an air quality forecasting activity with Anggraini, Lilis; Noersasongko, Edi; Marjuni, Aris; and Purwanto as activity participants. The activity explored multivariate hourly air quality forecasting using Multivariate Exponential Smoothing–Long Short-Term Memory (MES-LSTM) as the primary forecasting framework. Its relevance to Sustainable Development Goal 3 is reflected in the stated purpose of supporting environmental management, protecting public health, and enabling timely early-warning interventions.
The activity used the Beijing Air Quality dataset to predict carbon monoxide (CO), nitric oxide (NO), nitrogen dioxide (NO2), ozone (O3), and sulfur dioxide (SO2). A standardized evaluation protocol was applied across the models, using identical preprocessing procedures, chronological data partitioning, a 24-hour lookback window, and a one-hour forecasting horizon. MES-LSTM was evaluated against Long Short-Term Memory (LSTM), Temporal Fusion Transformer (TFT), and a hybrid MES-LSTM–TFT architecture.
The findings show that MES-LSTM achieved the lowest forecasting errors for several pollutants, particularly CO, NO2, and O3, while remaining competitive for NO and SO2. The hybrid MES-LSTM–TFT model did not improve forecasting performance and generally produced larger prediction errors than the corresponding single-stage models. Robustness analyses included five-seed experiments, persistence baselines, and multivariate coupling ablations. These analyses indicate that the effectiveness of pollutant interaction modeling depends on the target pollutant. They also assessed stability and reproducibility. Overall, the activity indicates that combining exponential smoothing with residual learning provides an effective and robust approach for short-term multivariate air quality forecasting, while greater architectural complexity does not necessarily improve predictive performance. In the SDG 3 context, the work provides forecasting evidence connected to public-health protection and early-warning support.