Lumping of atmospheric organic chemical species by machine learning

Lumping of atmospheric chemical species into different groups is one of the effective techniques used to reduce the complexity of the reaction mechanisms. Since lumping of chemical species into different categories is a classification problem, the application of machine learning by Artificial Neural Networks (ANNs) is appropriate to address the problem from a computational perspective. The conventional notation used to represent chemical species is not in a form which can be directly given as an input for machine learning. Issues such as what type of chemical information is appropriate and how best it is given as an input for ANN to obtain good results in classifying the chemical species into different lumped categories are discussed. Both the supervised and unsupervised learning methods are explored. The study in this thesis suggests that supervised ANNs can be gainfully employed for lumping of atmospheric chemical species when compared to the unsupervised ANNs.