Enhancing Customer Retention with SAP SAC Predictive Scenarios

In this newsletter, We'll delve into the practical usage and underlying principles of SAP Analytics Cloud's Predictive Scenarios. Delving into Predictive Analytics and Data Analysis may seem complicated, typically requiring advanced qualifications and technical know-how. However, SAP has revolutionized this perception through SAP Analytics Cloud (SAC) Predictive Scenarios. These tools empower business analysts and users by simplifying the process of generating meaningful insights from data, eliminating the need for manual algorithm crafting and data sifting. This article will explore SAC Predictive Scenarios and their functionalities.

What is SAP SAC Predictive Scenarios?

Machine Learning (ML) and AI are gaining popularity rapidly as businesses seek innovative solutions to meet their needs. SAP SAC Predictive Scenarios tool offers easy-to-implement classical ML methods with its user-friendly design, making it accessible for businesses looking to leverage these technologies. That tool designed to simplify the process of extracting valuable insights from data without requiring advanced technical skills. These scenarios utilize predictive analytics techniques to forecast future trends, identify patterns, and make data-driven decisions. With SAC Predictive Scenarios, users, including business analysts, can quickly create and analyze various types of data scenarios without the need for manual data sorting or complex algorithm development. This allows organizations to leverage their data more effectively to inform strategic decisions and drive business outcomes. SAC Predictive Scenarios offer three distinct types of data analytic scenarios below:

• Classifications
• Regressions
• Time Series Forecasts

Today we will explore the Classification feature of Predictive Scenarios with an example using the Bank churn dataset.

Data Description

Bank Customer Churn refers to the departure of customers or their transition to another bank. Anticipating churn through machine learning models empowers banks to proactively engage and implement strategies to retain customers, ensuring a stable and thriving customer community.

Classification

Classification involves sorting data into distinct categories or classes. In the context of predicting bank churn, the goal is to categorize customers into groups in a meaningful way. For instance, consider a dataset containing customer information such as demographics, transaction history, and account details. Classification would entail separating customers into those likely to churn and those likely to stay with the bank. This process helps identify how specific factors, such as account balance or transaction frequency, differ between churned and retained customers. In SAC Predictive Scenarios, the software analyzes the data using binary discrete variables, in our example it is the "Exited" column while considering other dataset attributes. These additional factors, known as "Influencer Contributions," reveal how various factors influence the classification outcome. Analysts can access individual statistics for each influencer, with SAC automatically generating these statistics upon training a classification scenario.

Example of Classification Feature

Import new data set as train and test.

In this dataset, we have 7999 rows and a total of 14 columns, consisting of 7 measures and 7 dimensions.

Before implementing the classification model, it's crucial to identify the columns, particularly the data and statistical types, to correctly utilize the model. Similar to traditional data preprocessing steps, checking for null values, outliers, and repetitive values is essential as part of Exploratory Data Analysis (EDA). SAC offers an intuitive UI that automatically calculates outliers, allowing users to apply statistical approaches with a simple click of a box.

Note: Outliers are extreme values within columns that can influence accuracy scores. However, despite this possibility, removing outliers may not always be the optimal solution.

As depicted in the image below, the details of the Age column indicate that beyond the age of 57, it becomes an outlier based on certain statistical calculations such as quartile values.

Furthermore, the data and statistical type of each column should be defined using the list box. Statistical types provide semantic context to column data, as elaborated below:

• Nominal: unordered, discrete values (for example, categories)
• Ordinal: sortable, discrete values (for example, rankings)
• Continuous: numerical, sortable, continuous values (for example, salaries, dates)

Textual: nominal values containing text (for example, sentences)

After all those data-processing steps, we can start employing our classification scenario.

1. Create a new classification model
2. Add training data source
3. Select your predictive goal
4. If needed, exclude the influencer columns, which are columns whose values influence the prediction. By default, all columns are considered potential influencers by the predictive model.
5. Click on the Train button.

In this example, we aim to predict the "Existed" column that shows the churn information of the bank customer. After training is completed, the accuracy scores of the model will be provided.

• Predictive Power serves as the primary indicator of predictive model precision. The closer its value is to 100%, the greater confidence you can have in the predictive model when generating predictions.
• Prediction Confidence reflects your predictive model's capability to maintain a consistent level of accuracy when applied to a new dataset with similar characteristics to the training dataset. Ideally, this value should approach 100%.
• The Gini index ranges from 0 to 1, with 0 representing perfect classification purity, where all elements belong to one class, and 1 indicating random distribution across classes. A value of 0.5 signifies an equal distribution of elements among classes.

Furthermore, during the training, Smart Predict calculates an optimized set of influencers to include in your predictive model.

We can assess the impact of various categories (individual values, value ranges, or value groups) on the target for each influencer. A higher absolute value of influence indicates a stronger effect on the category. The influence of a category may be either positive or negative. In this line, we can exclude influencer columns from our model to improve the accuracy. Then retrain and decide which is the optimal trained model for the prediction.

To compute our model to our test data, Test data should be added with the same data type.

• Choose Churn test dataset as the data source.
• Replicated Columns, by default only the columns that identify the entities of the data source are replicated. You can add another column to see in your result table.
• The statistics and predictions are calculated by the predictive model. It will be a column with the calculated prediction added to the generated dataset.
• Output as result_churn in the same folder. This will be our generated dataset.

Then click on the Apply button. It will create the predicted table by implementing the selected model to the test dataset.

Conclusion

In our churn analysis, we leveraged the key features of SAP SAC Predictive Scenarios using a machine-learning approach. The results are presented in a narrative format, offering valuable insights into the data to support informed decision-making. Feel free to explore our other newsletters for additional applications.

If you want to get more information to see if SAP Analytics Cloud is suitable for your requirements, we are happy to provide you with more insights with a live demonstration. If you want to learn more about our support and services, please visit our page.

Source: https://www.kaggle.com/code/kmalit/bank-customer-churn-prediction