Hybrid model for air quality prediction based on LSTM with random search and Bayesian optimization techniques

The state of the air-changing environment is a significant problem in urban areas that can cause serious health problems and economic performance. Accurate forecasting of air quality levels is a critical factor influencing public health and economic decisions. Accurate prediction of these variables can help with policy development and informed decision-making. This paper introduces a novel hybrid approach for air quality prediction, combining empirical mode decomposition, long short-term memory networks, and optimization techniques, namely random search and bayesian optimization. Empirical mode decomposition is used for decomposing the actual series into a subseries to reduce the data complexity and use long short-term memory (LSTM) networks to predict time series and employ a bayesian optimization and random search optimization approach to tune hyperparameters of LSTM. The hybrid model EMD-LSTM-Bayesian exhibits the lowest MAE value of 0.385 and the lowest RMSE value of 0.533, in contrast to the LSTM Model, which has the highest MAE value of 0.593 and the highest RMSE value of 0.804. The experimentation results suggest that the proposed hybrid method achieves higher accuracy as compared to other state-of-the-art methods. The percentage improvement of proposed hybrid model EMD-LSTM-Bayesian in terms of MAE%, 35.07, 20.94, 4.22, 27.35 for the comparison models LSTM, LSTM-Random Search, LSTM-Bayesian, and EMD-LSTM, respectively, for the dataset 2013 year.