Churn analysis with machine learning algorithms

Technology develops faster than the past in our time and tries to provide benefits in order to meet the increasing needs. Larger data are stored in electronic media, and the need to reach the stored data as quicker as possible is of great importance in an overcrowded population environment. Immense data are grouped by using data mining and classification algorithms which these data are labeled so that we can find in a very short time when necessary. The competition conditions are increasing rapidly in the logistics sector in a manner similar to nearly all developing sectors of today. The logistics industry, which is rapidly integrated by many developed countries, has started to deliver services in our country in the 1980s, and has been fully adapted in the 1990s. Customer loyalty for this fast-growing sector means tremendous importance all over the world. That is why greater emphasis should be placed on customer relationship management to minimize customer losses. With the increasing share of e-commerce sector, it is observed that more and more customer potential is in rise. At the same time, many new companies have joined the logistics sector. Under these conditions, it has become vitally important to retain existing customers and to identify customers who tend to move another companies. It has become the basis for standing in the sector to be able to understand the expectations of customers from the company, to follow up the company's attitudes better and accordingly to develop strategies in line with these findings. Late awareness about lost customers results in cost increases. New customer acquisition requires more time and more costs than keeping existing customers. Accordingly, the data obtained by taking into consideration the behaviors exhibited by the customers can be modeled and the customers who have a tendency to escape can be determined. Data mining is the activity of revealing meaningful information that is hidden in large data sets. Data mining can use both traditional statistical methods and machine learning methods. One of the most used areas is to estimate the customers that will be lost by segmenting the customers in customer loss analysis. In this study, we tried to reveal lost customer behaviors by analyzing the shipping information of past two years about customers of a logistics company operating in Turkey. It is applied on the support vector machine (SVM) algorithm which is one of the machine learning algorithms on the data prepared according to the principles of data mining. 2.000 customers among the data received from the company have been used in our application. Support vector machine algorithm, being the most suitable classification method, is used for this nonlinear data. The data for the past two years of these customers are divided into quarterly periods. Customer loss analysis was conducted on a total of 8 quarters and loss analysis estimation for the customers who tend to leave the company was made for the next three months.