Logistic evolutionary product-unit neural network classifier: the case of agrarian efficiency

Abstract

By using a high-variability sample of real agrarian enterprises previously classified into two classes (efficient and inefficient), a comparative study was carried out to demonstrate the classification accuracy of logistic regression algorithms based on evolutionary productunit neural networks. Data envelopment analysis considering variable returns-to-scale (BBC-DEA) was chosen to classify selected farms (220 olive tree farms in dry farming) as efficient or inefficient by using surveyed socio-economic variables (agrarian year 2000). Once the sample was grouped by BCC-DEA, easy-to-collect descriptive variables (concerning the farm and farmer) were then used as independent variables in order to find a quick and reliable alternative for classifying agrarian enterprises as efficient or inefficient according to their technical efficiency. Results showed that our proposal is very promising for the classification of complex structures (farms).