This paper concerns a method of selecting a subset of features for asequential logit model. Tanaka and Nakagawa (2014) proposed a mixed integerquadratic optimization formulation for solving the problem based on a quadraticapproximation of the logistic loss function. However, since there is asignificant gap between the logistic loss function and its quadraticapproximation, their formulation may fail to find a good subset of features. Toovercome this drawback, we apply a piecewise-linear approximation to thelogistic loss function. Accordingly, we frame the feature subset selectionproblem of minimizing an information criterion as a mixed integer linearoptimization problem. The computational results demonstrate that ourpiecewise-linear approximation approach found a better subset of features thanthe quadratic approximation approach.
展开▼
