Comparison of time series and machine learning (ML) models for pinapple price forecasting in Kerala

Abstract

The research work entitled “Comparison of time series and Machine Learning (ML) models for pineapple price forecasting in Kerala” was conducted at the College of Agriculture, Vellayani, Thiruvananthapuram, Kerala, from 2022 to 2024. The objectives of the study were to develop time series and Machine Learning (ML) models for price forecasting and comparison to determine the best performing model in pineapple price forecasting. Pineapple (Ananas comosus, Bromileacea) is one of the commercially important fruit crops of India, driven by it's popularity, nutritional benefits, and adaptability. Daily price data of both green and ripe pineapple for the past 18 years starting from January 2006 to December 2023 were collected from the website of Vazhakulam Pineapple Growers Association (VGA), Kerala. A comprehensive preliminary analysis was done in the collected data, which revealed significant autocorrelation, seasonality, and volatility in the same. In order to understand the behaviour of the time series, various statistical tests, such as those for normality and stationarity, were conducted. The data was pre-processed to handle missing values before model development. The dataset was then divided into training and test sets, with 80 percent of the data allocated for training the models and 20 percent reserved for testing their predictive performance. Several forecasting models were developed which include traditional time series models such as Autoregressive Integrated Moving Average (ARIMA), Generalized Auto-Regressive Conditional Heteroscedasticity (GARCH) and machine learning models such as General Regression Neural Network (GRNN), Support Vector Regression (SVR), Long Short-Term Memory (LSTM) networks and Facebook Prophet. Parameter tuning was done to optimize the hyperparameters of a model to improve its performance and to achieve the best possible accuracy for each of the models that were developed. The performance of these models was evaluated using multiple error metrics, including Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). These metrics provided a comprehensive view of the models' accuracy and robustness in forecasting pineapple prices. The results revealed that the LSTM model consistently outperformed all other models in terms of predictive accuracy, exhibiting the lowest error rates across all performance metrics. This superior performance can be attributed to LSTM’s ability to model long-term dependencies and capture intricate temporal patterns in the price data. Models like SVR and GRNN also showed better performance when compared to ARIMA and GARCH. On the other hand, the ARIMA model, while useful for short term predictions, displayed lower accuracy compared to other models, likely due to its limitations in handling non-linearities and complex seasonal patterns. The comparative analysis of the different models shows the effectiveness of machine learning techniques over traditional time series methods in price forecasting.