Abstract for: Machine Learning and Structure Representation for Crop Price Prediction
Classical statistical methods cannot deal with non-linear, complex functional relationships, nor with collinearity between variables. Most academic agriculture research in developing countries, fail to utilize easy-to-understand/teach modelling/analytic tools that can be disseminated to special stakeholders like rural farmers. In this paper we present an analysis of crop price predictions based on Artificial Neural Networks (ANN) and compare the results to linear regression methods and the insights generated from using systems thinking approach to understanding the crop price movements. Two reinforcing and two balancing feedback loops were identified via a collaborative loop diagram development with some farmers who were mostly able to comprehend the modelling process. Further development of this research will include developing a stock and flow model and the implementation of a recurrent neural network based on a larger dataset.