Forecasting Customer Churn with Machine Learning
As organisations migrate from a product-centric stance to more of a customer-centric orientation. Retaining existing customers, has become one of the key mantras for success.
Customer churn management can be a useful tool to gain competitive advantage, within industries, where the ability to switch from one brand to the other is relatively easy.
Hence, it is vital, that marketing and sales departments, incumbent, within organisations, that engage in high volume, low-value transactions.
Deploy data analytics to forecast customer churn. As this enables marketing and sales, to intervene, and reverse likely behaviours of consumers, planning to leave.
Through launching initiatives to mitigate the risk of customer churn and improve customer loyalty, retention and satisfaction.
Customer relationship management (CRM), is a strategy deployed by organisations to capture new customers, strengthen loyal and long-lasting customer relationships.
CRM techniques have been set-up across a broad array of industry sectors such as eCommerce, telecommunications, banking, insurance and retail.
Why Customer Leave Companies
Machine Learning Techniques Utilised for Customer Churn Prediction
There are several machine learning techniques the companies can deploy to forecast customer churn, such as artificial neural networks. One popular supervised model is the Multi-layer Perception trained with variations of the Back- Propagation algorithm (BPN).
So, what is back-propagation? Well, it is the central mechanism by which a neural network learns. Other techniques include;
Support vector machines: which are supervised learning models associated with learning algorithms. That analyse data and recognise patterns. They are used for classification and regression analysis.
Logistic Regression Analysis: This is a statistical process for estimating relationships, among variables. There are many techniques, for modelling and analysing several variables.
When the focus is on the relationship between the dependent and independent variable. Logistic regression analysis is a type of probabilistic statistical classification model that can be used to address churn prediction.
Analysing Customer Churn Using Microsoft Azure
Listed below is the output for the churn analysis of an anonymised bank. Over 10,000 records were analysed. The variables consisted of the following:
Results
Initially, we needed to train and test the data set. Hence the data was split two ways; 75% represents the trained data set and 25% represents the test data set.
The model used was the two-class logistics regression model, to predict customer churn.
In order to test the model, a score model module was selected to test 25% of the data in order to determine the performance of the model.
An evaluation module in Microsoft Azure was used to determine the accuracy of the overall results. That are listed below;
As you can see the model has an 82% accuracy rate. The lift curve is relatively high so the model apparently performed well.
True Positive (107) = for 107 records the model accurately predicted that the customers would churn.
False Negative (387) =for 387 records the model predicted that the customer churned when the customer did not churn.
False Positive (67) = for 67 records the model predicted that the customer did not churned when the customer did churn.
False Negative (1939) = for 1939 records the model predicted the customer would not churn and this was correct.
