Churn Prediction Using SAP Analytics Cloud

[rangeshtkb]

Introduction:

A predictive scenario is one in which we use machine learning, statistical modeling, and other data mining approaches to create predictions about future outcomes based on present data. When it's necessary to predict or discover future hazards, opportunities, or other requirements by looking for patterns in data, a predictive scenario is used. In general, a Predictive Scenario is a work area where we create predictive models and compare them to determine which one is the best. I have explored the SAP analytics cloud and its predictive capabilities under the guidance of akash_aj.

Predictive Scenario in SAC (SAP Analytics Cloud)

The following pieces make up a predictive scenario in SAP analytics cloud, which is formed around a modeling context:

• Target


• Type of an algorithm


• Filters


• Source of data

Use Case on Customer churn prediction in telecom

In developed countries, telecommunication has become one of the most important industries. The degree of competition has risen as a result of technological advancements and an increase in the number of operators. Companies are relying on a variety of techniques to thrive in this competitive industry. To increase revenue, three basic ways have been presented.

1. Attract new customers


2. Up sell to current customers


3. Increase the length of time clients stay with you.

To implement the third method, businesses must reduce the risk of customer churn, sometimes known as "moving from one provider to another". Customer churn is a significant issue and one of the most pressing challenges for large businesses. Companies are working to create methods to predict prospective customer churn because it has such a direct impact on their revenues, particularly in the telecom industry. As a result, identifying factors that contribute to customer churn is critical in order to take the required steps to reduce churn.

This blog post's main contribution is to demonstrate SAP analytics cloud's predictive capabilities by creating a churn prediction model in Predictive Scenario that helps telecom operators predict consumers who are most likely to churn. The model created in this blog post applies Classification Algorithms to Historical Data and creates a new way of engineering and selecting features.Many studies have shown that Predictive Analysis Technology is extremely effective at foreseeing this situation. Learning from historical data is used to implement this strategy.

1.Selecting a Datasets

The first step in predictive scenario is to identify the necessary data set. Here, a telecom company's 9 month historical data of customers who stayed and switched companies based on their usage records are taken for analysis.

SAP Analytics Cloud supports the data set in the following formats:

Before landing into Predictive Scenario in SAC, we need to model our data set in SAC. The prerequisite to run a predictive forecast in SAC is to enable planning options while creating a model, i.e., the planning model needs to be created to use predictive features. The data source can include flat files (CSV or excel) and acquired systems (BW, HANA). Live data source connection is not yet supported to enable the planning option. The modelling includes selecting and planning the required measures and dimensions and selecting date fields etc.

After Completing Modelling, it’s time for us to jump deep into predictive scenarios

1.Go to Predictive Scenario from the Menu Panel.

2.Select the required algorithms, SAC provides 3 machine learning algorithms. For our use case, we select the Classification algorithm and enter the Name and description for the model.

3.From the Settings Panel on the Right side, select your Training Model

4.After selecting your Training Model, Select your Predictive goal – target. Target is a column in the training data set which represent the output of the model in the historical data. Based on the data our model will predict the future output, in our case Churn column is our target.

5.Select the Influencer. Influencer column is nothing but a column in your training model which has the direct impact on the output. In our scenario the influencer columns are the user’s Recharge Plans, daily minutes spoken and ISD subscription etc.

6.Time to train our model – After selecting target and influencers, click train button to train your model.

7.After training is completed, you can see your model predictive power and confidence.

8.Influencer Contributions explains how many percent your individual influencer contributes to creating a model. Influencers have direct impact on the power and confidence of the model.

9.After Successful training, now it’s time to apply the model for testing.

Select the data source and enter the output name and path and click apply.

 

10.The predictive model being applied can be monitored like below.



 

11.After excluding these influencers Total_day_calls ,total_day_charge we can see that our models predictive Confidence increased drastically.

 



Comparing Accuracy of Both the Models

Comparing Accuracy of First Model

Comparing Accuracy of Second Model

The Accuracy of both the models can be found below

Conclusion:

Our Updated predictive model predicts the telecom churn with the prediction power 81.90% and prediction confidence 93.30%. I hope that this blog post would have clearly made you understand the SAC Classification predictive model with a practical use case. We have also seen  the importance of  proper influencers in deciding the model accuracy.

 

References:

Dateset from https://www.kaggle.com/c/customer-churn-prediction-2020/overview