An interpretable classifier with linear discriminant analysis based on AFS theory

There are many classification tasks in the real world and the digital world. Therefore, classification tasks have always been a hot spot. Nowadays, there are numerous classifiers, including traditional and intelligent methods. Most researchers focus on the accuracy and speed of learned models, they neglected the interpretability and comprehension of the model. In other words, most learned models are like black boxes, i.e., they are not interpretable. However, comprehension and interpretability are very important for a model. To make model interpretable and comprehensible, this paper proposes a new classifier which contains linear discriminant analysis (LDA) feature extraction based on Axiomatic Fuzzy Set (AFS) theory. The proposed method not only get a good performance in the accuracy rate but also has the capability of interpretability and comprehension. Because the proposed method can explore the class by giving semantic descriptions.