A Comparison of Time Series and Machine Learning Approaches for Forecasting Weekly Price of Garbled Black Pepper

Abstract

The present study has made an attempt to identify the best forecasting model to predict weekly price of garbled black pepper from January, 2000 to December, 2020, in Kochi market of Kerala, India. The volatility in prices of black pepper throughout the year poses a significant challenge for both farmers and consumers, being a perennial crop. Understanding the predictability of these price fluctuations in the near future is crucial for devising relevant policy recommendations. Consequently, price forecasting of black pepper is of paramount importance. Both time series and machine learning models have been used to forecast weekly prices of garbled black pepper. Models like Seasonal Autoregressive Moving Average model (SARIMA), Time-delay Neural Network (TDNN) model, and Long  Short-Term Memory model (LSTM) have been tried in the study to forecast weekly garbled black pepper price series. Based on the accuracy measures of the models fitted, TDNN (13:8s:1l) was the best model for forecasting the weekly price of garbled black pepper, for Kochi market of Kerala.