Comparative study of fuzzy methods applied to data classification.
The classification allows us to predict the value of a categorical attribute (discrete or nominal); it is a type of supervised learning since the cases of the set are labeled with the corresponding class. The construction of a learning model is based on the data set called the training set that will be used for the classification of new data. Fuzzy Logic (FL) based methods have been less applied in classification tasks that machine learning (ML) methods. FL is well recognized by its potential usefulness to knowledge formal representation and handling of uncertainty, thus simulating human knowledge capacity. The objective of this study is to compare two classification methods: an Artificial Neural Network (ANN) based on the K Nearest Neighbours (KNN) model and the Eureka Universe (EU) system. The first is an ML method, while the second is based on the use of Compensatory Fuzzy Logic (CFL), which is a type of FL. In this work, the EU system is compared against ANN-KNN on precision and interpretability criteria. The experimental analysis shows that ANN-KNN is highly precise but without the ability to interpret, while the precision of the EU depends on the type of used predicate but is interpretable. In the future, we seek the integration of these two learning models to complement each other and to have efficient and interpretable results. This kind of integrations is still moderate for FL and null to CFL.