Performance Comparison Of Classiers For Prospective Buyers Identication In Ethio Telecom Mobile Cross-selling Market

Direct marketing is a form of communicating an offer directly to a targeted group ofrncustomers through a variety of media. It plays a major role in customer retention andrnservice provisioning tasks. Retaining customers by providing products and services thatrnmeet their need is one of the main objectives of customer relationship management. Identifyingrnprospective customers for direct marketing enables a company to reach specificrnaudiences which will more effectively respond to promotions. Moreover, direct marketingrnhelps businesses to optimize their marketing budget, keeps current customers loyalrnto them, and makes businesses capable of measuring the result obtained from promotions.rnEthio telecom promotes service packages to its customers through SMS and mass communicationrnchannels. However, promotions should target customers based on the specificrnservices they use, and customer over-touching should be reduced especially duringrnSMS advertisement. In the current practice, no scientific methodology is implementedrnto estimate the potential respondents to cross-selling market promotion. Promotions arerncommunicated to both potential buyers and non-buyers without distinguishing the tworngroups. Direct marketing approaches help the company to effectively allocate resourcesrnand give services based on the interests of customers.rnThe aim of this thesis is to identify prospective customers in ethio telecom mobile valueaddedrnservice market. To achieve this goal, five classifiers namely Naive Bayes, Neuralrnnetwork, SVM, K-nearest neighbour, and Decision tree (J48) tested with customersrnservice usage historical data. In this process, 900,000 customers’ actual CDRs fromrnethio telecom were gathered and raw data aggregated with the aim of representing users’rnbehaviour. The representation was based on users’ responses towards service fee andrntime preference to use services. Sixteen feature variables and one predictor variablernare constructed from the raw CDR collected. Data cleaning and class balancing done,rnand the selected classifiers tested for their accuracy in identifying prospective buyers ofrnservice packages.rnThe right customers for direct marketing are identified and ways to minimize customerrnover-touching during the promotion of low-price packages are shown. So, beyond selectingrna classifier with high accuracy, the study aims at maximizing correctly classifiedrninstances. Accordingly, except Naïve Bayes classifier, the other four classifiers resultedrnin better performance. with Neural network classifier more than 90% of voice and 93%rnof SMS packages potential buyers are identified. whereas Decision tree (J48) has scoredrnthe best result with data package buyers dataset by identifying more than 94% of potentialrnbuyers.

Get Full Work

Report copyright infringement or plagiarism