Time Series Forecasting Using Parallel Randomized Fuzzy Cognitive Maps and Reservoir Computing

Fuzzy Cognitive Maps (FCMs) have been widely employed as nonlinear forecasting methods that are easily interpretable. They have a remarkable capability to enhance accuracy and are well-equipped to handle uncertainty and emulate the dynamics of complex systems. The main goal of this article is to present a new randomized multiple-input multiple-output (MIMO) FCM-based forecasting technique named M-PRFCM to forecast real-world high-dimensional time series in Internet of Things (IoT) applications. M-PRFCM is a first-order forecasting method integrating the concepts of randomized FCMs, Echo State Networks (ESNs), and Kernel Principle Components Analysis (KPCA). The training process of M-PRFCM is accelerated as a result of utilizing the ESN weight initialization trick, which randomly selects weights. The obtained results show the efficacy and validation of the proposed technique in terms of accuracy when compared with other existing approaches.