Substructural Surrogates for Learning Decomposable Classification Problems

摘要: This paper presents a learning methodology based on substructural classification model to solve decomposable problems. The proposed method consists of three important components: (1) structural model, which represents salient interactions between attributes for given data, (2) surrogate provides functional approximation the output as function attributes, and (3) predicts class new inputs. is used infer form surrogate. Its coefficients are estimated using linear regression methods. uses maximally-accurate, least-complex predict that yields an optimal searched iterative greedy search heuristic. Results show successfully detects interacting variables in hierarchical problems, groups them linkages groups, builds maximally accurate models. initial results non-trivial test problems indicate holds promise also shed light several improvements enhance capabilities method.