CO₂ concentration forecasting in smart cities using a hybrid ARIMA-TFT model on multivariate time series IoT data

Area of Science:

• Environmental Science
• Data Science
• Machine Learning

Background:

• Rising atmospheric Carbon Dioxide (CO2) levels drive global warming, with record highs observed despite economic slowdowns.
• Smart city initiatives leverage Internet of Things (IoT) technology for data-driven environmental management and emission reduction.
• Accurate CO2 forecasting is crucial for developing effective climate change mitigation strategies.

Purpose of the Study:

• To develop and evaluate a hybrid machine learning system for forecasting CO2 concentrations in smart city environments.
• To compare the proposed system's performance against traditional time series and deep learning models.
• To ensure the developed system is interpretable and provides insights into prediction drivers.

Main Methods:

• A hybrid machine learning system was developed using a multivariate time series dataset from IoT sensors measuring CO2 and environmental factors.
• The system's forecasting performance was evaluated against traditional time series methods and state-of-the-art deep learning architectures (e.g., transformers).
• Empirical comparisons were conducted across various scenarios, use cases, and forecasting horizons.

Main Results:

• The hybrid machine learning system demonstrated superior performance and interpretability compared to existing methods.
• Deep learning approaches generally outperformed traditional time series methods, especially for longer forecasting horizons.
• Statistically significant performance improvements were observed for the hybrid solution across different metrics and settings.

Conclusions:

• The developed hybrid system offers a highly accurate and transparent solution for CO2 concentration forecasting in smart cities.
• The findings highlight the potential of advanced machine learning techniques, particularly deep learning, in addressing climate change challenges.
• The system's interpretability provides valuable insights for understanding and managing urban carbon emissions effectively.