A hybrid prediction model integrating Fuzzy Cognitive Maps with Support Vector Machines

Abstract

This paper introduces a new hybrid prediction model combining Fuzzy Cognitive Maps (FCM) and Support Vector Machines (SVM) to increase accuracy. The proposed model first uses the FCM part to discover correlation patterns and interrelationships that exist between data variables and form a single latent variable. It then feeds this variable to the SVM part to improve prediction capabilities. The efficacy of the hybrid model is demonstrated through its application on two different problem domains. The experimental results show that the proposed model is better than the traditional SVM model and also outperforms other widely used supervised machine-learning techniques like Weighted k-NN, Linear Discrimination Analysis and Classification Trees.