Function Approximation Model Ensembles for the Simultaneous Determinations of Odor Classes and Concentrations

This paper presents a kind of combinative function approximation models to simultaneously estimate many kinds of odor classes and concentrations. A single approximation model, namely an expert, may be a multivariate logarithmic regression (MVLR), a quadratic multivariate logarithmic regression (QMVLR), a multilayer perceptron (MLP), or a support vector machine (SVM). An ensemble is made up of four such experts, and simulates the behaviors of gas sensor array to a specified kind of odor. The real outputs of each ensemble are the average predicted concentrations as well as relative standard deviations (RSDs) of odor samples. The ensemble with the most identical views finally gives the label and concentration of a sample. The "most identical view" is weighed by the sizes of RSDs given by all ensembles. The experimental results for 4 kinds of fragrant materials, 21 concentrations in total, show that the proposed approximation model ensembles and combination strategies are effective for simultaneously estimating many kinds of odor classes and concentrations