Neural Architecture Search Using DyNNO for Air Pollution Forecasting

The main objective of this study is to demonstrate how DyNNO, a neural architecture search algorithm based on multilayer perceptrons, can be applied to air pollution prediction using historical data from meteorological and environmental sensors. Data from various variables, including CO, NO, NOx, NO2, O3, PM10, PM2.5, RH, SO2, TMP, WDR, and WSP, have been collected and will be used for training DyNNO. Additionally, DyNNO’s results will be compared with other machine learning methods: Random Forest, Grid Search MLP, and Long Short Term Memory (LSTM). The results show that while DyNNO can generate optimal results with proper preprocessing, it still requires improvements to reach its full potential.