New Optimizer Improves Air Quality Forecasting with Neural Networks

Mary Joy Daniel Vinas· August 25, 2026 View original

Key takeaways

  • The QHAdamW optimizer significantly enhances ANN performance for air quality forecasting.
  • It improves model convergence, generalization, and accuracy compared to standard Adam.
  • The model can provide more reliable predictions for particulate matter, aiding environmental management.
  • Optimized optimizers are crucial for developing robust and accurate deep learning applications.

Who benefits

Environmental MonitoringPublic HealthUrban PlanningGovernment

Summary

This study introduces QHAdamW, an optimized Adam algorithm combining Quasi-Hyperbolic Momentum and decoupled weight decay, to enhance Artificial Neural Networks for air quality index forecasting. The proposed optimizer significantly improves convergence, generalization, and accuracy compared to the standard Adam optimizer.

Researchers have developed an enhanced optimization algorithm, QHAdamW, to improve the performance of Artificial Neural Networks (ANNs) in forecasting air quality. This new optimizer integrates the benefits of Quasi-Hyperbolic Momentum (QHAdam) and Adam with decoupled weight decay (AdamW), both of which are extensions of the widely used Adam optimizer. The goal was to address common issues like convergence speed, generalization ability, and overall forecasting accuracy in existing models. The study focused on predicting the Air Quality Index (AQI) for PM2.5 and PM10 using real-time data from Manila, Philippines, where the standard Adam algorithm is currently the only available model. Through extensive hyperparameter tuning, optimal values for QHAdamW were identified, leading to superior generalization performance. Comparative analysis using seven evaluation metrics demonstrated that the QHAdamW-enhanced model achieved lower error values and a regression coefficient closer to 1, indicating significantly improved accuracy. Furthermore, the model exhibited faster and more stable convergence, evidenced by lower loss values during both training and validation phases. This advancement offers a robust tool for environmental agencies, such as the Department of Environment and Natural Resources-Environmental Monitoring Bureau (DENR-EMB), to implement more effective and comprehensive air quality management strategies based on more accurate particulate matter forecasts.

Why it matters

Improved forecasting accuracy for critical environmental data like air quality directly impacts public health and policy decisions, offering a more reliable basis for interventions and resource allocation.

How to implement this in your domain

  1. 1Experiment with the QHAdamW optimizer in existing deep learning models for time-series forecasting or classification tasks.
  2. 2Benchmark the performance of QHAdamW against other state-of-the-art optimizers like Adam, AdamW, or SGD with momentum for specific use cases.
  3. 3Integrate enhanced forecasting models into environmental monitoring systems to provide more accurate and timely predictions.
  4. 4Collaborate with data scientists to fine-tune hyperparameters for QHAdamW to achieve optimal performance in specific datasets.

Original post by Mary Joy Daniel Vinas

"arXiv:2608.21463v1 Announce Type: new Abstract: The study employed an Artificial Neural Network in combination with the optimized Adaptive Moment Estimation (Adam) algorithm, currently the only AQI forecasting model available in the Philippines. The modified QHAdamW - Quasi-Hyper…"

View on X

Originally posted by Mary Joy Daniel Vinas on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses