Optimized Feature Extraction and Actionable Knowledge Discovery For Customer Relationship Management

Abstract

Recent years are envisaging intense competition in banking sectors and as a consequence, majority of banks are paying more attention on Customer Loyalty Prediction Customer Loyalty is considered as a major issue in Customer Relationship Management and is one of the most important and helpful analysis task used by telecommunication industries to maintain customers This study designs and develops techniques to improve the process of loyalty customer identification and also proposes techniques for action discovery For this purpose initially as a preprocessing task the missing values and outliers were removed to obtain cleaned data For customer loyalty assessment clustering and classification techniques were combined For customer loyalty assessment the customers were first classified into nonchurners and churners and then using clustering algorithm the nonchurners were further grouped as low medium and high risk customers A hybrid clustering model combining SOM Kmeans and DBSCAN was proposed using which the process of classification was enhanced For classification three classifiers namely SVM BPNN and Decision Tree classifiers were used newlineThe action discovery models first created customer profile using decision trees which were then used for action discovery The proposed model integrated data mining and decision making step together Curse of dimensionality was handled using techniques like Ant Colony Optimization or Unlimited Discriminant Analysis The customer profiles build with the help of decision tree learning algorithm was used to predict customer status Finally an optimized search for action was performed Experimental results showed that all the enhanced operations were successful and produced improved results when compared to the existing models newline