Predicting Churn of Credit Card Customers Using Machine Learning and AutoML

Introduction

In the last few weeks, 10% of credit card customers of a particular credit company churned from 55 different branches in 19 locations. Due to this, the Bank manager is in a dilemma because if he provides one specific scheme, some part of the customer goes out of service, while if he gives another service improvement, then other sections are affected. Nowadays, a major concern for most retail banks is the risk that originates from customer fluctuation, increasing the cost of almost every financial product. To control that, it becomes mandatory to know the number of customers who cease using the company’s service or product. This makes churn prediction the premiere to predict the number of customers most likely to churn shortly.

By doing so, we can take necessary steps to prevent them from churning by retaining them with better programs and facilities and creating special advertising through different media platforms. Data analysis on finance is more valuable because it has high leverage. One good finance decision creates a very good zero-cost replication effect. Data analysis gives you insight and continuously guides businesses to make the right decisions. Visualization of that data analysis also helps explain things more clearly. Nowadays AI is so popular and frequently used in every domain like business, medical, industries etc (R. Gupta et al., 2022, N. Pathik et al., 2022; S. Bharti et al., 2022; Yanping Zhao et al., 2022). In this work, we compared different approaches and algorithms to predict the relevant features that affect customer churn, which means we can find ways to reduce customer churn and create financial inclusion. Every bank has different requirements because the allocation of resources and their business operations are at different scales. In that case, we need a comparison of different methods and continuous improvement of our decision-making model. In advance, he needs a fixed analysis of the customer for different branches and churns prediction (Anil Kumar et al., 2008; Pamina et al., 2019) to understand various problems and give particular advertisements and facilities. To get this service company needs real-time data from customers and previous data records. We need to clean the data by preprocessing it to get reliable data.

Many algorithms exist for different applications since every algorithm has its resource requirements. According to requirements and practical applications, we set algorithms in the system and change it, or the system changes it automatically. One of the main goals for governments of developing nations is to increase financial inclusion and get as many people into formal banking solutions. For example, farmers take informal loans from local people with very high-interest rates and go into a debt cycle. We need a better-automated prediction model that makes good decisions and reduces our loan default to keep credit costs low. Every improvement makes the system more powerful since finance modelling is a high-leverage industry. When decisions made by machines are applied over and over again will give you a multiplier effect at the same cost. We did implementations of many different algorithms and compared them with each other. Different levels of business require different levels of accuracy and resources, so we give them one point solution to use our platform and make their finance process go very fast. In the entire system, we can also make this finance round automatic which allocates new money according to real-time data of customers. Contributions to the proposed work are: