Customer Churn Prediction for Financial Institutions Using Deep Learning Artificial Neural Networks in Zimbabwe

1.1 Background of the Study

According to Ahmed et al (2017), customer churn prediction models done through machine learning techniques gained massive popularity in just a few decades ago. Similarly, so as the trend in the financial industry where predictions have been done to predict dissatisfied customers who were likely to go for other service providers primarily using traditional machine learning techniques such as Support Vector Machine (SMV) and Decision Trees (DTs). Globally, Farquad, Vadlamani, Ravi, Bapi and Raju (2014) states that the Germany N26 Bank and American Express from America successfully implemented customer churn prediction. The American Express now relies on sophisticated predictive modelling that forecasts and prevents customer churn. This was achieved through the analysis of past customer transactions hence identification of customer accounts that are likely going to close and take preventive action to mitigate the problem. Amuda and Adeyemo (2019), argues that the cost of retaining existing customers is lower than the cost of acquiring new customers when they did a customer churn prediction using multi-layer perceptron for financial institutions in Nigeria. In the African continent, International Financial Corporation (IFC) World Bank (2018) states that Greenfield Microfinance in Sub-Sahara Africa is a business model to advanced financial inclusion which successfully implemented churn predictions. The following was its customer churn rate in the year of implementation.

Figure 1.

Greenfield microfinance churn rate

(adopted from IFC World Bank (2018))

This led to a 43 percent of adults in this continent to have access of formal financial services in comparison with a 23 percent as per the year 2011. It also states that its subsidiaries in South Africa managed to incorporate customer churn prediction which led to a successful avoidance of customer attrition. In Zimbambe, customer churn prediction has been done for mobile network operators by Kusotera, Chimire and Mapuwei (2018) in their study titled An Analysis of Customer Churn for Mobile Network Operators in Zimbabwe. Upon observation no bank has engaged in customer attrition predictions yet. This study aims at developing a customer churn prediction model using a case of one of the leading financial institutions in Zimbabwe.

1.2 Statement of the Problem

According to Amuda and Adeyemo (2019) tones and tones of data is being generated as a result of technological advancement hence data has become oil of the twenty first century however oil is just useless unless it is refined into fuel. Several techniques and methods were introduced for hidden meaningful and salient information to be extracted from this data. The techniques include traditional machine learning and deep learning artificial neural networks. Customer churn is a critical challenge for corporates as this entails loss of customers to competitors. An in advance prediction of customer churning behaviour is therefore crucial for high valuable insight extraction in order to retain and maximise a corporate’s customer base. In Zimbabwe, many if not all financial institutions have not yet implemented customer churn prediction hence failure to realise the benefits of customer retention at the expense of acquiring new customers. Therefore the purpose of this study is to develop a deep learning customer churn prediction model using customer past transactional data.