IFC-Filter

摘要: We survey the literature on fuzzy classification.We introduce a novel algorithm for membership function generation from data.We present software prototype that implements this algorithm.We evaluate our system qualitatively based two case studies in real organizations.Conclusion: resulting models can visually explain data to human experts. Fuzzy classification be defined as method of computing degrees objects classes. There are many approaches classification, most which generate sophisticated multivariate classify all input space simultaneously. In contrast, methods (MFG) derive simple map one variable class; therefore, by minimizing complexity, these very understandable The unique contribution paper is inductive logic. Most existing MFG apply either parameter optimization heuristics or unsupervised learning and clustering definition function. contrast heuristic methods, approximate functions any shape. comparison clustering, approach make use target signal learn supervised association between variables. Compared probabilistic translate frequency information, i.e., normalized histograms, directly into degrees, applies reasoning conditional relative frequencies, called likelihoods. According law likelihood logic, it ratio likelihoods interest when evaluating alternative hypotheses, not themselves. greatest advantage its understandability users thereby potential visual analytics. However, experimental evaluation did show reproducible significant effects predictive performance conventional regression models. Given there already accurate practical implication IFC-Filter unfold mainly explaining data, specifically, associations analytical variables, decision makers. Lessons learned with industry partners demonstrate extract interpretable actionable knowledge data.