Continuously self-adjusting fuzzy cognitive map with semi-autonomous concepts

Fuzzy cognitive maps (FCMs) are distributed computation systems used for qualitative modelling and behaviour simulation. Constructing an FCM is a time-consuming process and the quality of the resulting map is difficult to assess. In this paper we propose an extension to FCMs that self-adjusts the FCM based on real data from the modelled system. The self-adjusting FCM (SAFCM) changes the cause effect relationships and concept inferences for each system data point with the goal of reducing the error between real data and values produced by the map. In this way, the burden of map construction imposed on the map builder is reduced and the initially constructed map can be evaluated by examining the degree of change caused by the self-adjustment. We tested the SAFCM on two case studies where we measured the degree of change to the initial map structure set up by an expert. The experiments showed that the self-adjusted maps produced results that were closer to real data than the maps that were initially set up by the expert. We also compared the SAFCM to a basic FCM and to an FCM that used a standard learning algorithm. The results showed that our algorithm had higher accuracy.