Data-driven Modelling of Engineering Systems with Small Data, a Comparative Study of Artificial Intelligence Techniques

This paper equitably compares five different Artificial Intelligence (AI) models and a linear model to tackle two real-world engineering data-driven modelling problems with small number of experimental data. Analysis of results show that, in both cases, the models are highly nonlinear and Multi-Layer Perceptrons (MLPs) outperform other AI models including neuro-fuzzy networks (or in short fuzzy models), Radial Basis Function Networks (RBFNs) and Fully Connected Cascade (FCC) networks. The latter has been claimed to be superior in the literature for some non-engineering benchmarks.