Empirical Comparison of Fuzzy Cognitive Maps and Dynamic Rule-based Fuzzy Cognitive Maps

Among the soft computing techniques that can be used effectively to model decision tasks in autonomous robotics are Fuzzy Cognitive Maps. Dynamic Rule-based Fuzzy Cognitive Maps (DRBFCMs) are a Fuzzy Cognitive Map variant that allows modeling of dynamic causal maps, where influence weights are determined dynamically at simulation time using Fuzzy Inference Systems, in order to adapt to new conditions. We aim in this work to compare and contrast DRBFCM to a conventional Fuzzy Cognitive Map in application of cotton yield in precision farming. The cotton yield model shows the relationships between soil properties like pH, K, P, Mg, N, Ca, Na and cotton yield. DRBFCM was evaluated for 360 cases measured for three years (2001, 2003 and 2006) in a 5 ha experimental cotton field. The results revealed an accuracy of predictions of 85.55%, 87.22% and 73.33%, against 73.80%, 67.20% and 69.65% for the conventional FCM model, and against 75.55%, 68.86% and 71.32% for the FCM model with the Nonlinear Hebbian Learning algorithm, for the years 2001, 2003 and 2006 respectively. DRBFCM proved, in this case study, to predict more accurately the yield while being faithful to the real world model.