Feature Selection Based on Hierarchical Concept Model Using Formal Concept Analysis

Feature selection is used to reduce the number of input variables when using the predictive model. The hierarchical concept model is one approach for feature selection that exploit dependency relationships among hierarchically structured features. Thus, this work proposes the feature selection based on a hierarchical concept model using formal concept analysis. The high level that includes a set of attributes of structure will be selected because these high levels indicate general knowledge that represents some of the properties in child nodes. We experiment with selecting the top three high levels of hierarchical structure. Likewise, we compare the proposed model with the decision tree approach based on the hierarchical concept model using the top three high levels. The classification task is used to test the proposed model using 10 data sets from the UCI machine learning repository. The result shows that both hierarchical concept models are not different for classification performance.