Bioluminescence is the light emitted by a bioluminescent organism, produced by energy released from chemical reactions occurring inside the organism. The main focus in this study is to predict the expected number of bioluminescence sources in each depth in deep sea, using Bayesian regression trees (BART). Data for this analysis consists of number of bioluminescence sources and the depth of the sea it was recorded. BART is mostly used for higher prediction purposes and to determine the distribution of a species where the distribution over a spatial area does not seem continuous. BART is an efficient method to use when we have complex Bayesian hierarchical structures with uncertainty. The BART is a summation of nonparametric regression trees with priors to regularize the parameters in the tree model. Metropolis algorithm in MCMC is used to get the posterior distribution of the parameters where the distribution is unknown and the full conditionals are used for other known parameters. Tree prior was defined using the probability of a split in each terminal node and the probability of assigning splitting rules for each node. BART package in R-Studio was mainly used for this analysis and the results for the expected number of bioluminescence sources were obtained for each spitted groups of the explanatory variable. Here in this analysis, splitting rules are assigned as equally spaced across the explanatory variable. The resulting credible intervals showed an over fit in the model and this may be because the number of nodes in each tree is not controlled.
NarmadhaMM/BART-for-Bioluminescence
Predict the expected number of bioluminescence sources in each depth in deep-sea, using Bayesian additive regression trees (BART)
R