Short-term forecasting of vegetable prices based on LSTM model-Evidence from Beijing's vegetable data

Area of Science:

• Agricultural Economics
• Time Series Analysis
• Machine Learning

Background:

• Accurate vegetable price forecasting is essential for the agricultural sector's economic stability.
• Nonlinear patterns in vegetable pricing challenge conventional time series methods.
• The vegetable sector is a vital component of the national economy.

Purpose of the Study:

• To develop and evaluate a Long Short-Term Memory (LSTM) model for accurate vegetable price forecasting.
• To compare the LSTM model's performance against other machine learning techniques.
• To provide a tool for informed decision-making in the vegetable market.

Main Methods:

• Utilized average daily price data for six vegetable types from seven Beijing wholesale markets (2009-2023).
• Trained and tested a Long Short-Term Memory (LSTM) neural network model.
• Compared LSTM performance with Convolutional Neural Network (CNN)-based time series forecasting and other machine learning methods.

Main Results:

• The LSTM model demonstrated exceptional performance on the test dataset with an R2 score of 0.958 and MAE of 0.143.
• Achieved over 5% higher forecast accuracy compared to conventional machine learning counterparts.
• Exhibited robust predictive performance and generalization abilities across different vegetable categories.

Conclusions:

• The LSTM model offers a highly accurate and reliable approach to vegetable price forecasting.
• This forecasting tool can enhance market transparency, optimize supply chains, and support sustainable development.
• Applications include aiding growers, consumers, and policymakers in market decision-making.