Initial Work Towards a Framework for Timely Prediction of Merchant Churn in Informal Markets in the Real World with Small Quantities of Unbalanced Data in order to Influence Behaviour

South Africa has a a high percentage (over 80%) of pre-paid cell phone users. These users need easy access to commercial entities which can allow them to buy airtime and data vouchers. In South Africa, most major supermarkets, department stores, banks, and petrol stations sell such vouchers. To augment this, however, there is a growing population of informal traders which also sell such vouchers via apps on low cost Android point-of-sale (POS) devices. These Android POS devices connect to one of a handful of companies which aggregate vouchers from the various cell phone providers. This handful of companies is highly competitive and informal merchants often churn between the companies. This paper looks at the use of neural networks to predict whether or not a merchant is going to churn. This prediction gives the company time to offer the merchant some incentive not to migrate to the competition. Although there is a rich collection on literature on predicting merchant churn, this research is done specifically in the informal market with small quantities of data.