Combining Measures of Signal Complexity and Machine Learning for Time Series Analyis: A Review

Area of Science:

• Data Science
• Machine Learning
• Time Series Analysis

Background:

• Signal complexity measures (Hurst exponent, fractal dimension, Lyapunov exponents) quantify time series properties like persistence and predictability.
• These measures are crucial for understanding data characteristics relevant to time series prediction.

Purpose of the Study:

• To review complexity and entropy measures in conjunction with machine learning for time series analysis.
• To evaluate the application of these concepts in improving machine and deep learning predictions.

Main Methods:

• Comprehensive literature review of publications combining signal complexity measures with machine learning.
• Evaluation of existing applications and potential benefits for time series prediction and analysis.

Main Results:

• Signal complexity measures offer valuable insights into time series data, aiding in feature selection for machine learning.
• Incorporating complexity features can enhance the performance of machine and deep learning models, especially those with long-term memory.

Conclusions:

• Fractal and complexity measures can significantly improve existing machine and deep learning approaches for time series.
• Six distinct methods for combining machine learning with signal complexity measures were identified in the literature.