Relevance Vector Machines-Based Time Series Prediction for Incomplete Training Dataset: Two Comparative Approaches.

Area of Science:

• Machine Learning
• Time Series Analysis
• Data Science

Background:

• Supervised machine learning methods struggle with real-world time series containing missing data points.
• Accurate time series prediction is crucial across various scientific and industrial domains.

Purpose of the Study:

• To propose a novel time series prediction method for incomplete training datasets using relevance vector machines.
• To develop strategies for effectively handling missing inputs and outputs in time series data.

Main Methods:

• Utilized phase space reconstruction to establish relationships between missing inputs and outputs.
• Developed two imputation strategies for missing outputs: Expectation-Maximization and marginal likelihood maximization.
• Updated kernel matrix elements by imputing missing inputs with corresponding missing outputs.

Main Results:

• The proposed relevance vector machine-based method demonstrated robust performance on incomplete time series.
• Both proposed imputation strategies showed superior results compared to existing methods on synthetic and real-world datasets.
• The method successfully handles missing data, enabling direct modeling of incomplete time series.

Conclusions:

• The novel approach provides an effective solution for time series prediction with incomplete training data.
• The imputation strategies enhance the capability of relevance vector machines in handling missing data.
• This method offers a significant advancement for analyzing and predicting real-world time series data.