Interpretable Functional Logistic Regression

In this paper(1) we introduce a new approach, called "Interpretable functional logistic regression"(IFLR), which classifies functional data into two distinct groups and provides an easy-to-interpret and high predictive classifier. Our method is wide enough to handle different notions, though interpretability is comprehended in different ways. We regularize various derivatives of the coefficient function, so the coefficient function can be sparse, constant-wise, linear-wise or quadratic-wise, etc. In theory, we derive the asymptotic rate of convergence in our estimate. The usefulness of the IFLR approach is shown in the numeric world and we obtain interpretable classifiers with competitive and predictive ability.