Causality, Machine Learning, and Feature Selection: A Survey.

Area of Science:

• Data Science
• Machine Learning
• Causality Studies

Background:

• Understanding cause-and-effect relationships is crucial for complex data analysis.
• Traditional methods often rely on correlation, potentially missing vital causal links.
• Causality is key to advancing machine learning model robustness and accuracy.

Purpose of the Study:

• To review methods in causal discovery and causal inference.
• To highlight the application of causality in feature selection for machine learning.
• To demonstrate how causal reasoning improves decision-making in complex systems.

Main Methods:

• Review of causal discovery techniques for graphical representation of variable influence.
• Review of causal inference methods for quantifying variable impact.
• Exploration of causality-driven feature selection, particularly for sensor data.

Main Results:

• Causal reasoning enhances machine learning model performance in prediction and classification.
• Causality-based feature selection identifies critical links missed by correlation methods.
• Improved feature selection supports applications like fault and anomaly detection.

Conclusions:

• Integrating causal discovery and inference strengthens machine learning models.
• Causality-driven feature selection leads to more insightful and actionable results.
• This approach enables better decision-making in critical system maintenance and analysis.