Font Size: 

In the telecommunication industry, customer is the most valuable asset for the company to sustain the business process. The Moving of customers is certainly not expected so that companies need to analyze the customers profile to facilitate the conduct of business segmentation in support of decision making, the right decision is expected to generate value that could improve the performance and profit of the company. In analyzing customer data, past data is absolutely necessary that can describe the customer value in the future, whether the customer is potentially churn or non churn. In this case, using Support Vector Machine method for classification.

 

The SVM method was chosen because of its high accuracy in classification prediction. From this test will formed a performance that describes the results of the classification performance of churn and non churn. After the churn and non churn customer categories were obtained, the analysis of each variable of both billing historical and customer variables was obtained, it will be found that variables significantly effect and unaffected to the churn rate by reviewing the distance between performance with predefined threshold.

 

With this research, it would be expected to assist telecommunication company in conducting analysis of product focus and maintain customer loyalty that finally giving impact on sustainability and improvement of company performance.

Figure 1. Data Processing.

In Figure 1, it is explained that from each attribute in the tested data, consist of training data and testing data. Both data will be tested with SVM modeling in Figure 2 for getting performance that would be used for knowledge analysis.

 

Figure 2. SVM Model.

In figure 2, the threshold is determined at 5% of the total performance, if the attribute performance tested is smaller than the threshold, then the attribute has a significant effect, otherwise if greater than 5%, then the attribute does not significantly influence.

 

Table 1. Research Result.

From table 1 it can be concluded that there are 2 attributes that significantly influence the churn rate of total 7 attribute which have been tested. Usage Data in kb and Voice in Minutes where both attributes have a performance value smaller than 5% of the total performance of the overall attribute. While the attribute is approaching the significant in SMS that is equal to 9%. From these results, the telecommunications companies XYZ should maintain Data and Voice services in minimizing churn rate.

Data Mining, Support Vector Machines, Telecommunication, churn rate, significant.