Structural structure of the STDR-MNN Hybrid Model with Application

Abstract

Time series are always preoccupied with maintaining their stability to make precise forecasts about the future. So, a plethora of statistical models are emerged, by some focusing on mean stability and others on variance stability. Neural networks formation that resemble the structure of the nervous system in humans subsequently followed this. Now, in order to break down the time series into its constituent parts, our hybrid model will integrate the time series’ structure. the neural network performs the sub-network prediction procedure after receiving the divided sub-series. These predictions are then concatenated to provide the original network prediction. Time series data display a wide range of patterns. It is frequently helpful to dissect a time series into its trend, seasonality, cyclic variation, and irregular components in order to identify underlying patterns. The time series model is in charge of classifying the data series into four patterns such as: seasonality, trend, dispersion, and the remainder which are expressed in the model by S, T, D, and R respectively. By this work, we will introduce a hybrid model called STDR-MNN that combines a time series with a neural network. The four segmented series are sent to the MNN neural network, which uses them to forecast each segmented series individually. Finally, we will use the MATLAB2022A program to test a realistic application on the generated hybrid model in order to gauge its effectiveness with Real data.