Optimized PM2.5 Predictive model using Time Series by Genetic Algorithm Based Long Short Term Memory Networks

Abstract

Air pollution prediction model makes available of reliable evidence to take preventive measures for safeguarding country from severe pollutions. Air pollutants concentrations are simulated with the parameters, such as PM_2.5, CO, NO, SO₂, PM₁₀, rainfall, temperature, air pressure, humidity, and so forth. The long short term memory network is a specific type of Recurrent Neural Network majorly used in times-series data prediction. The dataset is collected from real-time air quality monitoring system maintained by Central Pollution Control Board (CPCB) through nationwide programs. Although, the LSTM model performs better always, still the hyper-parameter selection optimization is done manually. To address this concern, Genetic Algorithm based LSTM is proposed to develop the automated optimization of hyper-parameters for the prediction model to the major air pollutant concentration PM_2.5. The optimized model showed the greatest performance than the conventional models.