Hybrid Bi-Directional LSTM and Genetic Algorithm-Optimized Extreme Learning Machine for Air Quality Forecasting

Abstract

This study presents a robust and efficient framework for air quality forecasting that combines a Genetic Algorithm–optimized Kernel Extreme Learning Machine with a Bidirectional Long Short-Term Memory network to improve the prediction of PM2.5 concentrations. The Genetic Algorithm is utilized to fine-tune kernel parameters and optimize the hidden layer structure of the Extreme Learning Machine, which helps in selecting relevant features and reducing prediction errors. To better capture temporal dynamics, the Bidirectional LSTM processes time-series data in both forward and backward directions, allowing the model to understand dependencies from past as well as future observations. This dual-direction learning enhances the model’s ability to represent complex pollution trends and long-term patterns more effectively. The proposed hybrid approach is tested on real-world PM2.5 datasets and compared with traditional models such as standard Extreme Learning Machine and Support Vector Machine. The evaluation, based on metrics like Mean Squared Error and Root Mean Squared Error, shows a clear improvement in prediction accuracy and consistency. Overall, the results demonstrate that the proposed model offers a reliable and scalable solution for air quality prediction, making it suitable for applications such as real-time monitoring, early warning systems, and informed environmental planning.
Keywords—Air Quality Forecasting, PM2.5 Prediction, Bidirectional LSTM, Kernel Extreme Learning Machine, Genetic Algorithm Optimization, Hybrid Deep Learning Model, Time-Series Analysis, Environmental Monitoring, Machine Learning, Pollution Prediction, Feature Optimization, Predictive Modeling