Study of Rough Set Based Grey Relational BP Neural Network on Grain Yield Forecasting

Area of Science:

• Computational intelligence
• Data mining
• Machine learning

Background:

• Backpropagation (BP) neural networks are widely used for forecasting but suffer from imprecision due to redundant input nodes.
• High dimensionality in data can hinder efficient knowledge discovery and data mining processes.

Purpose of the Study:

• To enhance the prediction precision of BP neural networks by addressing input redundancy.
• To improve the efficiency of BP neural network training and reduce computational complexity.

Main Methods:

• Integration of rough set and grey relation theories to identify and reduce redundant input attributes.
• Utilizing grey correlation coefficients to determine attribute weights and refine the decision table.
• Training the BP neural network with the optimized, reduced set of condition attributes.

Main Results:

• Significant improvement in prediction precision for China's grain yields in 2001 (0.83%) and 2002 (1.93%).
• Achieved over 99% fitting precision for grain yield data from 1990-2000.
• Reduced the number of input and hidden nodes, leading to increased network training rates.

Conclusions:

• The proposed method effectively reduces dimensionality and eliminates redundancy in BP neural networks.
• This optimized model demonstrates a promising approach for accurate forecasting and efficient data mining.
• The integration of rough set and grey relation theories offers a robust strategy for improving predictive modeling.