Predictive modeling of haze using chaos theory and deep learning algorithms

With the swift growth of urbanization and industrialization, fine particulate matter (PM$_{10}$) has escalated into a major global environmental crisis.  PM$_{10}$ is often used as a haze indicator, severely affecting human health and ecosystem stability.  Accurate prediction of PM$_{10}$ levels is crucial, but existing models face challenges in handling vast data and achieving high accuracy.  This study investigates four years of PM$_{10}$ time series in industrial area in Malaysia.  Paper aims to develop and compare haze predicting models using chaos theory, including the local linear approximation method (LLAM), local mean approximation method (LMAM); also several deep learning algorithms including convolutional neural network (CNN) and bidirectional LSTM.  The performances of these models are evaluated using root mean square error (RMSE), correlation coefficient ($r$), and coefficient of determination ($R^2$).  The result shows that Bi-LSTM provides the highest accuracy, LLAM outperforms CNN, while LMAM is reliable but less precise.