Soft Computing-Based Ensemble Technique for WPI Estimation in India's Textile Sector

The current work presents a novel ensemble neuro-fuzzy forecasting approach for the wholesale price index (WPI) of textile commodities (a total of twenty-five items) of the present Indian WPI series. The offered technique used two softcomputing methods – an extreme learning machine (ELM) neural network and an Abbasov-Mamedova (AM) fuzzy timeseries (TS) model. The proposed ELM-AM uniquely fused the outcomes acquired from both elements (i.e., ELM and AM models) through a weighted averaging strategy to construct the final ensemble output. The present work employs two accuracy metrics, i.e., forecast-MAPE and forecast-RMSE, and multiple (five) forecast horizons, i.e., three, six, nine, twelve, and eighteen months horizons, to assess the proposed ELM-AM's forecasting ability. The proposed ELM-AM exhibited high accuracy for all the twenty-five items' WPIs across the first four forecast horizons, with high accuracy observed in twenty-two out of twenty-five, i.e., 88% of the cases in the remaining horizon. The proposed ELM-AM outmatched twenty-four diverse approaches (i.e., both component models, four automatic soft-computing techniques, six commonly used automated TS forecasting approaches, six state-of-art hybrid techniques, and six contemporary ANFIS models) and, thus, found to be a fit contender for forecasting the WPI of textile items in India.