Heterogeneous feature selection for classification of customer loyalty fast moving consumer goods (Case study: Instant noodle)

In the face of ASEAN open market, the actors (Fast Moving Consumer Goods industry) must increasingly explore patterns of business development because of tough competition and challenges in the market. One of strategy for surviving in high competition is retain the customer loyalty. The data were usually obtained from various sources and contains heterogeneous features, such as numerical and non-numerical features. The datasets with heterogeneous features can affect feature selection results that are not appropriate because of the difficulty of evaluating heterogeneous features concurrently. In this paper, we propose a method that combine the features transformation and subset selection based on mutual information to obtain feature subset that able to improve performance classification algorithm. Analysis comparative among before feature subset selection, dynamic mutual information (DMI) methods, p-value methods and researcher estimate were also done. Feature transformation (FT) is another way to handle the selection of heterogeneous features. The datasets were obtained from the survey of customers fast moving consumer goods, with a total of 386 respondents. By applying unsupervised feature transformation and dynamic mutual information methods, can be known the relevant features that affected the performance of decision tree algorithm. The accuracy and F-measure increased of the DMI-unsupervised-feature-transformation compared to all features (without features subset selection), p-value methods and manual features subset selection. The accuracy and F-measure for DMI-unsupervised-feature transformation are 76.68% and 73.5% respectively.