A Comparison of Classical Time Series Models and Machine Learning LSTM Model to Forecast Paddy Production in Sri Lanka

Abstract

The most common & effective traditional methods of univariate time series forecasting are Autoregressive Integrated Moving Average (ARIMA) family models and Exponential Smoothing family models. With the recent advancement in more advanced machine learning algorithms and approaches such as Long-Short-Term-Memory modeling approaches, new algorithms are developed to analyze and forecast time series data. The objective of the study is to identify the best time series forecasting model among classical time series models and machine learning LSTM model to forecast the annual paddy production of Sri Lanka. Based on the RMSE, MAE and MAPE values, the results showed that the estimated error of ARIMA & Double Exponential Smoothing (DES) models are much higher than the estimated error of chosen LSTM model. Hence LSTM model outperforms to the traditional-based algorithms like ARIMA and smoothing models for forecasting the paddy production of Sri Lanka. The forecasts for paddy production from 2022 to 2024 were 4.92, 4.89 and 5.34 million Mt respectively. This model can be used by researchers for forecasting paddy production in Sri Lanka and it should be updated continuously with incorporation of recent data.