Presenting a New Approach for Predicting and Preventing Active/Deliberate Customer Churn in Telecommunication Industry Majid Joudaki Islamic Azad University, Doroud Branch Doroud, Iran m.joudaki@gmail.com Mehdi Imani Islamic Azad University, Science and Research. Qazvin Branch Qazvin, Iran m.imani@gmail.com Maryam Esmaeili 17shahrivar Higher Education Center Karaj- Iran laleh_kimiya@yahoo.com Mahtab Mahmoodi 17shahrivar Higher Education Center Karaj- Iran mahtab_empire2@yahoo.com Niloofar Mazhari Allameh Dehkhoda Higher Education Institute Qazvin, Iran nl.mazhari@gmail.com Abstract - Like other storing and retrieval systems, the systems of telecommunication networks contain large databases which should store loads of data temporarily/permanently per moment. This data set includes data concerning customers and calls placed on the telecommunication networks. Extracting useful and relevant information buried under these vast telecommunication data sets is time-consuming and tiresome in emergency cases for considerations or finding occurred troubles a solution. In this paper, we firstly study the different techniques of data mining in some cases such as fraud detection, customer profiling and marketing. Then we consider the specified algorithms of fraud detection systems, creating a customer profile for distinguishing or classifying business and residential customers and how to increase the number of members of these networks in marketing. In the end, a new method of estimating time of customers' giving-up for any reason and consequently membership contract cancellation called "customer churn" is presented to avoid customer churn in telecommunication companies. Keywords - Fraud detection, Marketing, Customer profile, Customer Churn I. INTRODUCTION he telecommunication networks generate and store a large amount of data such as Call Detail Data, information on each call placed and Customer Data, specifications of each customer. Manual analysis of this great amount of data in the urgent circumstances is not impossible but too difficult to handle and for this reason we need a kind of systems to identify unusual manners or illegal process for instance the fraudulent phone calls on time. But unfortunately the required information of these systems must be obtained from human resources that it was time-consuming in many cases. It is expected that data mining techniques are able to remove all the existing problems related to the above-mentioned matter in the telecommunication industry. The below notes are considered in all the data mining applications studied in this paper: • Scale of data inserted in the records of telecommunications databases. • The raw data such as call detail data should be summarized by the useful summary features before effectively mined. • Real-time performance: any model / rule must be applied in real-time for instance a fraud detection. II. TYPE OF TELECOMMUNICATION DATA Before using data mining technology in any field, we should first understand which types of data will be evaluated. Two types of telecommunication data are necessary for the following applications and described completely in this section. A. Call Detail Data Information on the call placed on a telecommunication will be saved as call detail record. Telecommunication databases contain a lot of call detail records generated in real time and typically should be kept online. They have sufficient information like the originating and terminating phone number, date, time and duration of each call and etc. For describing customer’s behavior we need to combine these records with customer data and summarized in a single record. The features inserted in this summary will provide us with enough and short information on each customer immediately in case of need. Below is a sample of a customer’s profile based on the received/dialed calls in a month. Figure 1- A sample of customer’s profile We can profile residential and business customers easily according to item 5, 6, 9. According to item 9, telemarketing customers obviously call many different T