Discriminant analysis using the unweighted sum of binary variables: a comparison of model selection methods

Area of Science:

• Biostatistics
• Machine Learning
• Clinical Decision Support

Background:

• Many clinical decision rules utilize linear discriminant functions.
• These functions often simplify to the unweighted sum of binary variables (SBV).

Purpose of the Study:

• To propose and evaluate methods for forward stepwise selection of SBV models.
• To compare the performance of different SBV model selection strategies.

Main Methods:

• Development of several forward stepwise selection algorithms for SBV models.
• A simulation study was conducted to assess method performance under diverse conditions.

Main Results:

• No single SBV model selection method demonstrated uniform superiority for fixed-size models.
• Key factors influencing method performance include the sample size to variable ratio and data's class-conditional moment structure.

Conclusions:

• Practical strategies for constructing effective SBV models are offered.
• The choice of selection method should consider specific data properties and research goals.