A Novel Approach for Optimal Cluster Identification and N-Order Hesitation Based Time Series Forecasting

Time series forecasting is an important field of research, especially when the series is completely random, known as a strictly non-stationary time series (NS-TS). To handle the randomness efficiently, the paper presents a novel approach to identify the optimal number of clusters in NS-TS followed by a hesitation-based N-order IFS forecasting algorithm (called IFSH_new). The performance of the proposed approach for determining the optimal number of clusters is demonstrated using four variants of the proposed algorithm, IFSH_new, across five datasets. Furthermore, a comparison of the best variant of the proposed approach with existing state-of-art work based on mean absolute percentage error (MAPE) and root mean square error (RMSE) confirms the superiority of the proposed model over others. Additionally, the graphical representation of actual versus forecasted data points for different datasets shows good improvement at high change points compared to other approaches.