An innovative hybrid approach to forecasting İnterval Time Series data with Elman Artificial Neural Networks and a modified adaptive Network-Based Fuzzy Inference System

Interval valued Time Series (ITS) techniques have been employed to conduct both modeling and forecasting in the data analysis. This manuscript recommends a hybrid model that combines two operative methods, which are the Modified Adaptive Network Based Fuzzy Inference System (MANFIS) and Elman Artificial Neural Network (ERNN) model, to be employed for ITS to generate a forecast. The ITS mainly differs from conventional (non-interval) time series by taking into account both highs and lows of interval simultaneously. By doing so, possible interrelations between bounds can be taken into account. The recommended hybrid strategy takes into account this aspect of the data to forecasting interval time series data combining both ERNN and MANFIS. The recommended method composes of two parts. The initial part constructs the algorithm pertaining to the ANFIS structure. The second part is based on the ERNN model structure. The advantages of the proposed method can be expressed as in 1. The recommended method, called ERNN-MANFIS, uses the optimization method of particle swarm optimization to train the model, 2. Addressing both linear and nonlinear aspects of forecasting, this approach offers dual advantages, known respectively as model-based and data-based approaches. It is trained by using fuzzy c-means method and particle swarm optimization. In the fuzzification step of the input values, membership values are systematically obtained by using fuzzy c-means clustering technique. By doing this, the prediction performance of the method is improved. Seven different real datasets are used to affirm the effectiveness of the mentioned approach. In addition, the results were compared with the results of previous models available in the literature. In conclusion, comparisons have been provided to show the effectiveness of the recommended model.