Distributional regression modeling via generalized additive models for location, scale, and shape: An overview...

Area of Science:

• Statistics
• Machine Learning
• Educational Data Mining

Background:

• Technological advancements generate vast amounts of unstructured data in research.
• Learning analytics (LA) and educational data mining (EDM) utilize unsupervised machine learning (ML) for analyzing educational data.
• Existing methods often struggle with the complexity of educational datasets.

Purpose of the Study:

• To overview the power and flexibility of Generalized Additive Models for Location, Scale, and Shape (GAMLSS) in relation to ML techniques.
• To highlight GAMLSS's capability for causal inference through causal regularization.
• To demonstrate GAMLSS application in LA using a real-world dataset.

Main Methods:

• Overview of Generalized Additive Models for Location, Scale, and Shape (GAMLSS) as a supervised statistical learning framework.
• Comparison of GAMLSS with unsupervised machine learning (ML) algorithms commonly used in LA/EDM.
• Discussion of causal regularization for GAMLSS to enable causal inference.

Main Results:

• GAMLSS provides a flexible supervised framework for modeling all distributional parameters of response variables.
• GAMLSS demonstrates significant advantages over traditional ML techniques for LA/EDM tasks.
• The framework can be extended for causal analysis, offering deeper insights.

Conclusions:

• GAMLSS is a powerful and flexible supervised learning approach for LA and EDM.
• The GAMLSS framework offers enhanced capabilities for data analysis and causal inference in educational settings.
• This statistical approach provides a valuable alternative to unsupervised ML methods.