Future Challenges of Particulate Matters (PMs) Monitoring by Computing Associations Among Extracted Multimodal Features Applying Bayesian Network Approach

The particulate matter (PM) is emitted from diverse sources and affects the human health very badly. In the past, researchers applied different automated computational tools in the predication of PM. Accurate prediction of PM requires more relevant features and feature importance. In this research, we first extracted the multimodal features from time domain standard deviation average (SDAPM), standard deviation of standard deviation (SDSD), standard deviation of particulate matter (SDPM), root-mean square of standard deviation (RMSSD), and nonlinear dynamical measure wavelet entropy (WE) – Shannon, norm, threshold, multiscale entropy based on KD tree (MSEKD), and multiscale approximate entropy (MAEnt). We then applied the intelligent-based Bayesian inference approach to compute the strength of relationship among multimodal features. We also computed total incoming and outgoing forces between the features (nodes). The results reveal that there was a very highly significant correlation (p-value <0.05) between the selected nodes. The highest total force was yielded by WE-norm followed by SDAPM and SDPM. The association will further help to investigate that which extracted features are more positively or negatively correlated and associated with each other. The results revealed that the proposed methodology can further provide deeper insights into computing the association among the features.