Robust statistical methods for high-dimensional data, with applications in tribology.

Area of Science:

• Statistics
• Machine Learning
• Tribology

Background:

• Practical experimental data often contains outliers and high dimensionality (more variables than observations).
• Existing robust statistical methods are limited for high-dimensional data with outlying observations.
• Robust methods for low-dimensional data are well-developed but face challenges in high dimensions.

Purpose of the Study:

• To provide an overview of selected robust statistical methods for high-dimensional data.
• To demonstrate the application of these methods using real-world tribology datasets.
• To evaluate the effectiveness of robust methods in handling data with outlying observations.

Main Methods:

• Overview of robust regression and classification techniques suitable for high-dimensional data.
• Implementation of selected robust methods.
• Application and evaluation on two high-dimensional tribology datasets.
• Utilizing pre-processing and sampling strategies alongside robust methods.

Main Results:

• Robust statistical methods, when combined with appropriate pre-processing and sampling, improve prediction performance.
• These methods provide valuable insights into data subsets that deviate from the majority.
• The study demonstrates the practical applicability of robust methods in tribology.

Conclusions:

• Robust statistical methods are essential for analyzing challenging high-dimensional experimental data.
• Strategic pre-processing and sampling enhance the efficacy of robust methods.
• The application in tribology highlights the potential of these techniques for uncovering hidden data structures and improving model accuracy.