Comparing ARIMA, Neural Network and Hybrid Models for Forecasting Fish Production in Bangladesh Authors Maisha Binte Saif

Abstract

Time series forecasting is a commonly applied method for scientific predictions. There are several econometric methods for forecasting time series observations and predicting the systematic pattern of underlying data. ARIMA model is most renowned in this aspect for its linearity. Nowadays a machine learning model namely artificial neural network (ANN) is gaining popularity for its nonlinear characteristics. Inconsistent conclusions may frequently be drawn when evaluating whether ARIMA models or neural networks are better at forecasting future events. For this reason, a hybrid methodology has been established in this study to get advantage from both linear and nonlinear modeling. The annual dataset of fish production in Bangladesh from 1990 to 2020 has been evaluated in this case. Formulating three measurement errors, RMSE, MAE and MAPE it has been demonstrated that the hybrid approach has high level of forecasting accuracy than the other two models in forecasting fish production data.