Income prediction in the agrarian sector using product unit neural networks

Abstract

Abstract European Union financial subsidies in the agrarian sector are directly related to maintaining a sustainable farm income, so its determination using, for example, the farm gross margin is a basic element in agrarian programs for sustainable development. Using this tool, it is possible the identification of the agrarian structures that need financial support and to what extent it is needed. However, the process of farm gross margin determination is complicated and expensive because it is necessary to find the value of all the inputs consumed and outputs produced. Considering the circumstances mentioned, the objectives of this research were to: (1) select a representative and reduced set of easy-to-collect descriptive variables to estimate the gross margin of a group of olive-tree farms in Andalusia; (2) investigate if artificial neural network models (ANN) with two different types of basis functions (sigmoidal and product-units) could effectively predict the gross margin of olive-tree farms; (3) compare the effectiveness of multiple linear, quadratic and robust regression models versus ANN; and (4) validate the best mathematical model obtained for gross margin prediction by analysing realistic farm and farmer scenarios. Results from ANN models, specially the product-unit ones, have provided the most accurate gross margin predictions.