Pay Attention to Evolution: Time Series Forecasting with Deep Graph-Evolution Learning.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Deep Learning

Background:

• Traditional statistical and ensemble methods often underperform deep learning for time-series forecasting.
• Existing approaches frequently overlook the sequential nature and multivariate interdependencies within time-series data.

Purpose of the Study:

• To introduce a novel neural network architecture, Recurrent Graph Evolution Neural Network (ReGENN), for enhanced time-series forecasting.
• To address the limitations of existing methods by incorporating inter-temporal and intra-temporal relationships.

Main Methods:

• Developed a novel neural network architecture, ReGENN, integrating graph evolution with deep recurrent learning.
• Modeled temporal data dependencies considering both inner variables (intra-temporal) and outer variables (inter-temporal).
• Inferred multivariate relationships between co-occurring time series.

Main Results:

• ReGENN demonstrated significant performance improvements, achieving up to 64.87% over competing ensemble and statistical methods.
• Analysis of intermediate weights revealed that simultaneous consideration of inter- and intra-temporal relationships boosts forecasting accuracy.
• The method effectively captures how multiple multivariate data synchronously evolve.

Conclusions:

• ReGENN offers a powerful new approach to time-series forecasting by effectively modeling complex interdependencies.
• The findings highlight the importance of considering both internal and external temporal relationships for accurate multivariate time-series prediction.
• This work advances deep learning applications in artificial intelligence for sequential data analysis.