Flexible Bayesian Nonlinear Model Configuration

Regression models are used in a wide range of applications providing powerful scientific tool for researchers from different fields. Linear, or simple parametric, often not sufficient to describe complex relationships between input variables and response. Such can be better described through flexible approaches such as neural networks, but this results less interpretable potential overfitting. Alternatively, specific parametric nonlinear functions used, the specification is general complicated. In paper, we introduce approach construction selection highly regression models. Nonlinear features generated hierarchically, similarly deep learning, have additional flexibility on possible types considered. This flexibility, combined with variable selection, allows us find small set important thereby more Within space functions, Bayesian approach, introducing priors based their complexity, A genetically modi ed mode jumping Markov chain Monte Carlo algorithm adopted perform inference estimate posterior probabilities model averaging. various applications, illustrate how our obtain meaningful Additionally, compare its predictive performance several machine learning algorithms.