Tea Price Prediction Using Hybrid Arima-BP Neural Network Model

Abstract

This study examines the effectiveness of a hybrid ARIMA-BP Neural Network model in predicting global tea prices. The model integrates ARIMA to capture linear trends and Back Propagation Neural Network (BP NN) to address non-linear patterns, combining their strengths for improved forecasting accuracy. Using monthly tea price data from 2015 to 2022 obtained from Index Mundi, the study evaluates the hybrid model's performance against standalone ARIMA. Stationarity of the data was achieved through first-order differencing, and model selection was based on Akaike Information Criterion (AIC) and diagnostic checks. The hybrid model demonstrated superior predictive accuracy, achieving a Mean Absolute Percentage Error (MAPE) of 6.32%, compared to 12.79% for ARIMA alone. These results underscore the potential of hybrid models for volatile commodity markets, offering practical implications for risk management and decision-making in the tea industry. Stakeholders can leverage this model to anticipate price fluctuations, optimize operations, and enhance financial planning. Future research could explore hybrid approaches in other commodities and incorporate additional predictive variables.

Keywords: tea price prediction, ARIMA, back propagation neural network, hybrid model, forecasting, commodity markets.