Extending statistical boosting. An overview of recent methodological developments.

Area of Science:

• Statistical modeling
• Machine learning
• Biomedical research

Background:

• Boosting algorithms have gained significant attention for simultaneously estimating and selecting predictor effects in statistical models.
• Recent advancements in the last decade have focused on enhancing these algorithms.

Purpose of the Study:

• To review recent methodological developments in boosting algorithms for statistical modeling.
• To emphasize topics particularly relevant for biomedical research applications.

Main Methods:

• A unified framework is proposed for gradient boosting and likelihood-based boosting (statistical boosting).
• This approach integrates methods previously addressed separately in the literature.

Main Results:

• Methodological developments in statistical boosting over the past decade fall into three main categories.
• These include variable selection for sparser models, diverse predictor effect types, and extensions to new regression settings.

Conclusions:

• Statistical boosting algorithms now enable unbiased variable selection and automated model choice.
• These algorithms are adaptable to nearly any regression setting and various predictor effect types.