Hybrid Neural Network With K-Means For Forecasting Response Candidate In Direct Marketing

• Main Author: Ramadhan, Rakhmat Sani
• Format: Thesis
• Language: English; English
• Published: UTeM 2014
• Subjects: Q Science (General); QA76 Computer software
• Online Access: http://eprints.utem.edu.my/id/eprint/16255/1/Hybrid%20Neural%20Network%20With%20K-Means%20For%20Forecasting%20Response%20Candidate%20In%20Direct%20Marketing%2024%20Pages.pdf; http://eprints.utem.edu.my/id/eprint/16255/2/Hybrid%20Neural%20Network%20With%20K-Means%20For%20Forecasting%20Response%20Candidate%20In%20Direct%20Marketing.pdf; http://eprints.utem.edu.my/id/eprint/16255/; http://library.utem.edu.my:8000/elmu/index.jsp?module=webopac-d&action=fullDisplayRetriever.jsp&szMaterialNo=0000091008

The larger Bank‘s electronic data customer provides difficulty a marketing campaign. An efficient marketing campaign is needed to promote a product and services. The predictive data mining techniques use to help a marketing analyst provide more value to their customers by the right offer because of decreasing in responses to a direct marketing campaign. Distribution of customer data record in marketing response data are often found issue of imbalanced dataset. This study proposed hybrid Neural Network (NN) methods in data mining to support direct marketing analysis and forecast. Backpropagation NN is supervised learning methods that analyze data and recognize to solve many problems in the real world by building a model that is trained to perform well in some non-linear problems. K-means algorithm grouping process by minimizing the distance between the data and designed can handle very large dataset also continuous and categorical variable for handling imbalanced dataset. This research concerns on binary classification which is classified into two classes. Those classes are yes and no. The data was collected from the Machine Learning Repository Dataset in the University of California Irvine (UCI).This experiment compares hybrid K-Means + NN with basic NN. The result shows the improvement of accuracy from 91.53% to 91.59%, recall 22.15% to 27.7% and F-Measure 44.23% but not to precision from 61.69% to 60.75%.