Artificial neural networks used for investigation of fatty acid content of Romanian sunflower oilseeds genotypes

The objective of this study was to investigate the efficiency of Artificial Neural Networks (ANNs) in classifying and predicting the fatty acid content from Romanian sunflower oilseeds genotypes, as solutions of computational engineering problems. The two-layer probabilistic ANN, using a radial basis layer and a competitive layer, has been used for classification. There were two criteria of classification, the degree of polyunsaturation and the linoleic/oleic acid ratio, which allowed the defining of two categories. The first ANN has been designed for classifying the first category into three groups defined by the polyunsaturated fatty acid content: group 1 of less than 40% polyunsaturated fatty acid, group 2 of 40%-50% polyunsaturated fatty acid, and group 3 of higher than 50% polyunsaturated fatty acid. The classification was based on the following acids in the samples: C14:00, C15:00, C16:00, C16:01, C17:00, C18:00, C18:01, C18:02, C18:03, C20:00, C20:01 and C22:00. The second designed ANN has been used for classifying the category of linoleic/oleic acid ratio into three groups: group 1 of linoleic/oleic acid ratio higher than 2, group 2 of linoleic/oleic acid ratio between 1 and 2 and group 3 of linoleic/oleic acid ratio less than 1. The results of both classifications revealed a good accuracy of the trained ANNs for classifying the sunflower oilseeds. The numerical tests demonstrated the computational advantages of the prediction methodology.