Fuzzy Cognitive Maps and Multi-step Gradient Methods for Prediction: Applications to Electricity Consumption and Stock Exchange Returns

The paper focuses on the application of fuzzy cognitive map (FCM) with multi-step learning algorithms based on gradient method and Markov model of gradient for prediction tasks. Two datasets were selected for the implementation of the algorithms: real data of household electricity consumption and stock exchange returns that include Istanbul Stock Exchange returns. These data were used in learning and testing processes of the proposed FCM approaches. A comparative analysis of the two-stepmethod of Markov model of gradient, multi-step gradient method and one-step gradient method is performed in order to show the capabilities and effectiveness of each method and conclusions are based on the obtained MSE, RMSE, MAE and MAPE errors.