Two forecasting model selection methods based on time series image feature augmentation.

Area of Science:

• Agricultural Economics
• Data Mining
• Machine Learning

Background:

• Accurate forecasting of agricultural product prices is vital for market stability and agricultural data mining.
• Existing forecasting methods struggle with inefficient feature engineering and imbalanced datasets.

Purpose of the Study:

• To propose a novel predictive model selection approach for agricultural product price forecasting.
• To address challenges of imbalanced data and improve efficiency and accuracy in forecasting.

Main Methods:

• Time series data transformed into images using Gramian Angular Fields (GAF), Markov Transition Fields (MTF), and Recurrence Plots (RP).
• Information Fusion Feature Augmentation (IFFA) method to combine time series images (TSCI).
• Convolutional Neural Network (CNN) classifier for model selection, enhanced with Transfer Learning (TL) and S-Folder Cross Validation (S-FCV).

Main Results:

• The proposed IFFA-TSCI-CNN-SFCV method demonstrates superior performance compared to existing approaches.
• The method effectively handles imbalanced sample data and mitigates overfitting.
• Experimental results show significant improvements in both efficiency and accuracy of agricultural price forecasting.

Conclusions:

• The novel approach provides an effective solution for predictive model selection in agricultural price forecasting.
• The integration of time series image encoding and advanced machine learning techniques enhances forecasting capabilities.
• This method offers a robust framework for agricultural data mining and stream data event analysis.