Department of Biostatistics

People

Prospective Students

Academics

Research

News & Events

Calendar

Consulting

Employment Opportunities

Resource Quicklinks

Computing Environment

Contact

Logic Regression Methodology and Software

Ingo Ruczinski, Department of Biostatistics, Johns Hopkins University

Logic Regression is an adaptive regression methodology that attempts to construct predictors as Boolean combinations of binary covariates. In numerous situations when most predictors are binary, the interactions between predictors may be what causes a difference in response. This issue arises, for example, in the analysis of single nucleotide polymorphism (SNP) data and data mining problems. In the proposed methodology we create new predictors from binary predictors, generating rules of the form ``$X_1$, $X_2$, $X_3$ and $X_4$ are true'', or ``$X_5$ or $X_6$ but not $X_7$ are true''. These new predictors and their coefficients in a regression model are estimated simultaneously using a simulated annealing algorithm. The publically available software is organized such that the Logic Regression methodology can be applied to any (regression) model, as long as a score function can be defined. Such score functions were implemented for linear regression, logistic regression and Cox models, but a user can easily supply her/his own function, if for example a certain covariance structure needs to be modelled. I will describe the methodology and the software and give some examples.