������

The doctoral thesis of Zhiqiang Liao, MEng. “Overfitting reduction in convex regression: Theory, models, and applications” will be publicly examined at the Aalto University School of Business on Friday, September 5, 2025.

In the era of big data, overfitting is a critical challenge that must be addressed carefully. Data collected in the real world almost always contain random noise, measurement errors, or irrelevant variance. For example, when studying consumer preferences, responses may vary depending on mood or recent events unrelated to long-term preferences.

In convex regression, overfitting refers to the phenomenon where the statistical model fits not only the underlying relationship between variables but also random noise, measurement errors, or irrelevant variance. Due to its high flexibility, convex regression is more prone to overfitting than simple linear regression.

In this dissertation, I examine the overfitting problem in convex regression from three perspectives: theory, models, and applications. I find that overfitting often occurs in convex regression, especially near the boundary. Based on this theoretical finding, I propose different methods to reduce overfitting in convex regression. I further investigate their application in the regulation of electricity distribution networks in the Nordic countries.

To conclude, I contribute to the research on overfitting reduction in convex regression from theoretical, modeling, and practical angles.

Keywords: Convex regression, overfitting, regularization, optimization

Opponent: Professor Kristofer Månsson, Jönköping University
Custos (Chairperson): Professor Lauri Viitasaari, Aalto University School of Business. 

Link to the doctoral thesis:

Zoom link: https://aalto.zoom.us/j/64303430934

Contact information:
zhiqiang.liao@aalto.fi
+358 504673051