Comparison of Bayesian and Frequentist Neural Networks on Air Quality Time-Series Problems

This paper aims to compare Bayesian and frequentist versions of Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) neural networks via cross validation and bench-marking by addressing a regression problem on a time-series dataset using similar network structures. We compare the model performance using the coefficient of determination (R2) score, Pearson and Spearman correlation coefficients by varying size of the dense units in the recurrent layers in both deep learning models.<p></p>