Modified BBO-Based Multivariate Time-Series Prediction System With Feature Subset Selection and Model Parameter...

Area of Science:

• Artificial Intelligence
• Machine Learning
• Time-Series Analysis

Background:

• Multivariate time-series prediction is a complex challenge in data modeling.
• Echo State Networks (ESNs) are effective for time-series prediction but require optimal feature selection and parameter tuning.

Purpose of the Study:

• To propose a novel Modified Biogeography-based Optimization Echo State Network (MBBO-ESN) system.
• To simultaneously achieve optimal input feature subset selection and ESN model parameter optimization for improved prediction accuracy.

Main Methods:

• Developed an improved evolutionary algorithm, MBBO, incorporating an S-type migration model, covariance matrix migration, and Lévy mutation.
• Implemented a hybrid-metric feature selection method within MBBO-ESN to identify important input features.
• Optimized key ESN parameters using the MBBO algorithm.

Main Results:

• The MBBO algorithm demonstrated superior performance compared to traditional evolutionary algorithms.
• MBBO-ESN achieved more accurate multivariate time-series predictions on benchmark and real-world datasets.
• The proposed system automatically identifies feature-parameter relationships, enhancing predictive power.

Conclusions:

• MBBO-ESN offers a competitive and effective approach for multivariate time-series prediction.
• The MBBO algorithm enhances exploration and rotation invariance, improving optimization capabilities.
• The study highlights the significance of joint feature selection and parameter optimization for ESN performance.