Design Novel FCM-based Forecasting Model for Interval-valued Time Series: From Distance-metric Perspective

Predicting interval-valued time series (ITS) has long been a subject that has attracted researchers from a diverse range of fields including economics and finance. Since the data and relations in real world are usually highly sophisticated and inaccurate, modeling real complex systems is a challenging task especially for large scale, inaccurate and non stationary datasets. Fuzzy Cognitive Map (FCM) has become a powerful tool for modeling and analyzing complex systems, and been widely used in various fields. As the increasing use of FCM, there is still very little known about implications of selecting the fitness function during the weight training process. Therefore, the purpose of this study is to explore the effects of different distances on the accuracy of learning algorithm in FCM by conducting experiments on different datasets. Findings reveal that the choice of distance indeed impacts on the final performance of the model. Moreover, this paper employs the enhanced overlap distance as the fitness function to optimize the weights. Suitability of this distance has been demonstrated, showcasing a significant advantage in accurately calculating the intervals' distance.