Analysis of The Customer Churning on Digital Media Users with a Sample Data

The customer satisfaction has a huge impact on service delivery of any business. A simple word of mouth opinion structures the business environment to enhance their productivity and delivery. With such impact from customers, it is essential to keep them on track, to know the value of the product and service. The approach used for this project is to analyze the users of digital media, to check if they could continue the business with the organization, if not, make them to do business with help of more service delivery. For this analysis, a sample of digital media user’s data was considered, to know if they could possibly churn in future. This prediction was done with the help of machine learning techniques. The tool used for this analysis was Rapidminer. The output was shown with accurate results in statistical representation.

Introduction

In general, CRM (Customer Relationship Management) is a tool which helps organization to maintain the relationship between the buyers and customer’s interaction, track their records and accounts. It helps them to improve the customer satisfaction. For an analysis, a sample data of digital media was considered for churn prediction. This analysis is to predict whether if a customer would opt to stay with organization even after the contract period. This is similar to attrition model. Customer retention is an important aspect in in any organization, where it shows the level of company’s performance from low to high. The attrition is also one of the major employments of Data mining.

In current era, everything is becoming digital. The usage of digital media is becoming a necessity for the survival in business environment. This helps the organizations and customers to get updated on the trend for their own purposes. There are several forms of digital media in various formats such as audio, video, images and graphical representations. Considering the attrition model, there are three types namely voluntary attrition, involuntary attrition and expected attrition. If a customer wanted to switch to another company, it is voluntary attrition. Involuntary attrition also known as forced attrition is when the customer is terminated by the company for any reason, some common reasons are unpaid bills. Expected attrition is when the customer no longer is available in the target area, for instance when a customer moves to another place. There are multiple methods to predict the outcome of this project. The main background of this project is to look on Survival analysis. In this analysis, the machine learning techniques are employed to check the variation between them. They are Deep learning and Logistic regression. With the help of such techniques, the best accurate method will be known and can be taken for consideration. To perform this analysis, a tool called ‘Rapidminer’ was used.

Research and investigation of techniques

There are various techniques available to implement and get results from the prediction of churn analysis of customers in digital media. The techniques can be of a machine learning technique such as Bayesian network, Deep learning or decision trees. In other way, it can also be a statistical method of prediction through Logistic regression, which performs mainly between dependent variable and other variable when the dependent variable is dichotomous. There were some previous works which was done on this project with certain techniques. All those techniques gave mere output as expected. The dataset used for this project is very much balanced. This helps the ML techniques to perform analysis and give effective results. In case of Imbalanced, the techniques will not work and efficient results will not be available. However, for imbalanced datasets, there is a technique called Oversampling Technique, which deals with classification problems, has two types. They are Synthetic minority oversampling technique and Adaptive synthetic sampling technique. This technique helps in balancing the datasets, which helps in performing the analysis. Another popular technique used for Churn analysis is CART, which is Classification and Regression Tree model. This is the branch of Decision tree model. This technique mainly deals with classification and misclassification problems in the dataset. The other popular model for this analysis that was used was Support Vector Machine (SVM) model. This model also works mainly on classification linearity problems. It is effective in working on linear and non-linear cases. The above mentioned models are not limited, but were worth to mention on using for this churn analysis. It has a special way to apply on certain hypothesis to be more effective.

Techniques used

As discussed earlier, many important techniques are available in use. But in this project, only two techniques are used to find the churn analysis in digital media. These techniques are so popular and widely used for such kind of project in churn analysis. This technique helps us not only in predicting the outcome, but also helps us statistically with all factors that are leading for a customer to either stay or go for another network. The dataset used for this project has 21 columns. The column ‘Churn’ is the dependent variable. It is a dichotomous variable with yes or no. The Independent variables are Senior citizen, Gender, Tenure months, Phone service, Multiple lines, Internet service, Online security, Online backup, Device protection, Tech support, Streaming TV & Movies, Contract period, Paperless billing, Payment method, Monthly charges and Total charges.

A. Neural Networks (Deep Learning)

This is one of the popular algorithms, in the area of prediction analysis. It is one of the branches of machine learning techniques. This big data processing is able to analyze large amount of data at a particular time, however it may also take some amount of time to run the dataset if the data volume is very high. This technique is more flexible and scalable. The analysis was performed using the Rapidminer tool. In this test, accuracy is calculated with the overall variables. The metric type for this test is binominal. The confusion matrix algorithm is used for the statistical classification of the dataset. With the help of simulation, a deep understanding is analyzed with what sort of customer prefer amenities with bills they receive. To analyze the performance, tests such as Precision, AUC, sensitivity, specificity, recall, f measure and accuracy were performed.

B. Logistic Regression

This is also one of the methods of Machine learning techniques. This is the statistical method of prediction. This method could be the best technique for this project as it deals with customer attrition cases. This analysis is significant when the dependent variable is dichotomous. The output is coded as 0 or 1. Only binary classification is followed in this method. Logistic regression classified into binomial, ordinal or multinomial. This regression helps users in describing the data. It also helps in explaining the link between dichotomous variable and independent variable. The analysis was performed using the Rapidminer tool. In this model, the co-efficient, standard co efficient, standard error, z-value and p-value of each attributes were analyzed. There is a lift chart, where the relationship between target and population was examined. To check on complete performance, tests such as accuracy, AUC, sensitivity, specificity, recall, f measure and precision were performed.

Conclusion

Thus, we analyzed the customer churning on digital media users with a sample data. Several reasons were available as reason for a customer to switch service providers. To see a deep view on attrition, couple of data mining techniques was used, implemented the approach and the results were displayed. To understand the /technique, justification of the usage was also discussed. The analysis was performed using the Rapidminer tool. The tool helps to vary the output in the form of bars and graphs. Two important machine learning techniques were considered; they are Deep learning and Logistic regression. Logistic regression found out to be the best model for this analysis, with the help of values from accuracy and ROC curves. Since this model deals mostly with dependent variable, when it is dichotomous, it predicts and evaluates accurate results. From the models analyzed, it is said that customer attrition was caused mainly because of contract period and monthly subscriptions. To strengthen this analysis, few tests were made through logistic regression and it pinpoints key reasons where the customers might fall and leave the company. To overcome this solution, companies would need to re work on their subscription methods to retain customers and analyze them of sticking to the same company.