Hybridized Time Series Modeling with Machine Learning Techniques for Pomegranate Price Fluctuations

Abstract

Accurate and strategic price forecasting plays a crucial role in enabling stakeholders to make well-informed decisions and effectively navigate the complexities of market uncertainties. Price data is inherently complex, and conventional models often fall short in capturing its intricate patterns and variations. To address these challenges, models that can capture both linear and nonlinear relationships are necessary. Hybrid models have emerged as a robust solution to capture linear and nonlinear patterns in complex datasets, making them particularly effective at addressing the complexities of price prediction. This study highlights the efficiency of hybrid models in price modelling using the price data of pomegranate, which is recognized for its significant nutritional and economic importance in the horticultural sector. Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), Root Mean Square Percentage Error (RMSPE), and Median Absolute Percentage Error (Median APE) used to assess the model performance. The findings reveal that hybrid models, such as ARIMA-SVR in Bangalore, ARIMA-ANN in Chennai and Hyderabad, and ARIMA-RF in the Trivandrum market consistently outperform the benchmark models. These hybrid approaches combine the strengths of conventional and machine learning techniques, providing more reliable and accurate predictions for the distinct data characteristics of each region.