A Hybrid Approach for Forecasting Mustard Price having Long-Memory Property

For the modelling of time series data having long memory properties, we generally use the autoregressive fractionally integrated moving average (ARFIMA) model. This model performs well compared to the autoregressive integrated moving average (ARIMA) model. However, it cannot capture the nonlinear property of the data. In order to achieve the desired and accurate forecasts, hybridizing the existing forecasting models is an important technique. The hybrid time-series model combines the strength of individual models. Accordingly, this study proposes a hybrid model based on ARFIMA and extreme learning machine (ELM) for agricultural time-series data with long memory properties. For evaluation of the proposed model, the daily mustard price (₹/q) of Agra and Bharatpur markets from 1 January 2016 to 31 January 2020 was used. Empirical results show that the forecasting performance of the proposed hybrid model based on ARFIMA and ELM is better than the existing models.