Customers Churn Prediction and Attribute Selection in Telecom Industry Using Kernelized Extreme Learning Machine and Bat Algorithms
Journal: International Journal of Science and Research (IJSR) (Vol.5, No. 12)Publication Date: 2016-12-05
Authors : S. Induja; V. P. Eswaramurthy;
Page : 258-265
Keywords : Churn prediction; Expectation Maximization; Kernelized Extreme Learning Machine; data preparation; pre-processing; attribute selection; BAT algorithm; Naive Bayes Classifier;
Abstract
With the fast development of digital systems and concomitant information technologies, there is certainly an incipient spirit in the extensive overall economy to put together digital Customer Relationship Management (CRM) systems. This slanting is further more palpable in the telecommunications industry, in which businesses turn out to be increasingly digitalized. Customer churn prediction is a foremost aspect of a contemporary telecom CRM system. Churn prediction model leads the customer relationship management to retain the customers who will be possible to give up. Currently scenario, a lot of outfit and monitored classifiers and data mining techniques are employed to model the churn prediction in telecom. Within this paper, Kernelized Extreme Learning Machine (KELM) algorithm is proposed to categorize customer churn patterns in telecom industry. The primary strategy of proposed work is organized the data from telecommunication mobile customers dataset. The data preparation is conducted by using pre-processing with Expectation Maximization (EM) clustering algorithm. After that, customer churn behavior is examined by using Naive Bayes Classifier (NBC) in accordance with the four conditions like customer dissatisfaction (H1), switching costs (H2), service usage (H3) and customer status (H4). The attributes originate from call details and customer profiles which is enhanced the precision of customer churn prediction in the telecom industry. The attributes are measured using BAT algorithm and KELM algorithm used for churn prediction. The experimental results prove that proposed model is better than AdaBoost and Hybrid Support Vector Machine (HSVM) models in terms of the performance of ROC, sensitivity, specificity, accuracy and processing time.
Other Latest Articles
- Human Activity Detection using RGBD Images
- Nullity and Energy Bounds of Central Graph of Smith Graphs
- Synthesis, Characterization of Imide Substituted Polyphosphazenes and Evaluation of their Structural Properties
- Design and Implementation of 16 X 16 High speed Vedic multiplier using Brent Kung Adder
- Determinants of Unmet Need for Contraception among Currently Married Women in Oromia National Regional State; Evidence from Ethiopia Demographic and Health Survey Data
Last modified: 2021-07-01 14:48:53