Characteristic mango price forecasting using combined deep-learning optimization model.

Area of Science:

• Agricultural Economics
• Machine Learning
• Time Series Forecasting

Background:

• Accurate product price forecasting is crucial for scientific decision-making and industrial planning.
• Mango price prediction is economically significant due to the fruit's role in regional development.
• High volatility in mango prices presents significant forecasting challenges.

Purpose of the Study:

• To develop and evaluate a deep-learning combination forecasting model for mango prices.
• To improve the accuracy and reliability of mango price predictions.
• To support the fruit industry with better forecasting tools.

Main Methods:

• A combination forecasting model integrating back-propagation (BP) neural networks and long short-term memory (LSTM) networks was proposed.
• Daily mango price data from a major Chinese fruit wholesale market (January 2014 - April 2022) was utilized.
• The model learned and predicted mango price changes using historical data.

Main Results:

• The BP-LSTM model achieved a root-mean-square error of 0.0175 and a mean absolute percentage error of 0.14%.
• The R2 determination coefficient reached 0.9998, indicating a high degree of accuracy.
• The combined model outperformed individual BP and LSTM models in prediction accuracy and generalizability.

Conclusions:

• The proposed BP-LSTM deep-learning model offers superior performance for mango price forecasting compared to standalone models.
• The model demonstrates excellent accuracy, fitting the actual price trends effectively.
• This approach provides a valuable tool for enhancing decision-making in the fruit industry.