Application of the self-organizing map in the classification of natural antioxidants in commercial biodiesel

Abstract

The parameters relative protection factor, induction period, rate constant, density, acidity number, water content, flash point, viscosity, cloud point and pour point of 47 biodiesel samples containing the antioxidants of extracts of senna leaves, hibiscus flowers and blackberries were determined. The objective of this research was to apply the self-organizable map (SOM)-type network, using data on the physicochemical properties of biodiesel. SOM is a neural network built on a uni- or two-dimensional grid of neurons to capture the important characteristics in the data contained in a large amount of input. The results were tabulated and presented to the SOM neural network for the classification of antioxidants according to efficiency. A network with 35 × 35 topology was used for the segmentation of the samples. By analyzing the weight maps, it was possible to verify that the most adequate parameter in the classification was the relative protection factor. The analysis showed that two distinct clusters were formed, one for the senna extract and the other including extracts of blackberries and hibiscus flowers. This type of methodology proved to be effective in the selection of natural antioxidants according to regional availability by classification mainly according to the efficiency of protection of the oxidation process in biodiesel, since natural antioxidants with similar properties in the SOMs can be easily substituted for one another in order to minimize costs.