Integration of RNN with GARCH refined by whale optimization algorithm for yield forecasting: a hybrid machine learning approach

Forecasting yield is a challenging task in all agricultural crops. So, it is imperative to develop a machine learning hybrid model with available data for yield forecasting. The main objective of this research is to develop a novel hybrid model for forecasting the sugarcane yield on non-linear time series data. Recurrent neural network typically holds a long memory that allows sufficient forecasts with a fewer number of parameters. The weights and thresholds of the recurrent neural network are optimized by whale optimization algorithm to improve the efficiency of the neural network, and thus obtaining accurate results. Consecutively increasing the performance in forecasting volatility of time series is a challenging task. BPNN-GARCH, RNN-GARCH, and novel WOARNN-GARCH referred to as hybrid models; fitted with precipitation as well as sugarcane yield data and a novel method was found more appropriate for forecasting the volatility of rainfall as well as sugarcane yield for medium-term which facilitated in observing the future incidences for the next few years. Combining a statistical model like generalized autoregressive conditional heteroskedasticity with recurrent neural network refined by whale optimization algorithm grants a novelty of predicting the yield with volatility in time series analysis.