کشف مشتریان سودآور با رویکرد داده‌محور

هدف: امروزه مشتریان به عامل بسیار مهم و حیاتی در هدایت سرمایه گذاران، تولیدکنندگان و حتی محققان و نوآوران مبدل گشته‌ اند. به همین دلیل، سازمان‌ ها نیاز دارند مشتریان خود را بشناسند و برای آنان برنامه ‌ریزی کنند. در این پژوهش، تلاش شده تا به یکی از اساسی ‌ترین سوالات سازمان ‌های بیمه‌ ای، یعنی پیش ‌بینی سطح خسارت مشتریان، پاسخ داده شود.روش تحقیق: در پژوهش حاضر از ابزار داده‌ کاوی برای داده ‌های مشتریان صنعت بیمه، بخش بیمه بدنه خودرو از سال 1394 تا 1396 استفاده شده‌است. تعداد کل داده‌ها که از ابتدا در این پژوهش مورد استفاده قرار می‌گیرد بیش از 19356 بوده که در ادامه و در طی آماده‌سازی آن‌ها با استفاده از نرم‌افزار rapidminer 7.1 تعداد داده‌هایی که در نرم‌افزار لحاظ می‌شود 19356 است. پس از پردازش اولیه تلاش می‌شود، از بین 15 متغیر موجود در پایگاه داده ویژگی استخراج شود که ملموس باشد و این پژوهش را در هدف خود یاری دهد. بدین منظور با به کارگیری خوشه ‌بندی، رانندگان بر اساس میزان مبلغ خسارت به خوشه ‌های مجزا تقسیم می شوند و ویژگی ‌های هر خوشه بیان می ‌شود. در قسمت خوشه‌ بندی، ابتدا الگوریتم‌های k-means،  kmedoidsو dbscan استفاده شده‌ است. سپس الگوریتم‌ های بکار رفته به جهت زمان انجام محاسبات و میزان صحت با یکدیگر مقایسه شدند.یافته‌ها: در نهایت الگوریتم k-means به عنوان الگوریتم بهینه برای این مجموعه داده انتخاب شد. در انتها به کمک درخت تصمیم مدلی پیش بینی ارایه می شود که شرکت ‌های بیمه را در جهت سودآوری بیشتر و کشف مشتریان سودآور کمک می کند و برای برنامه‌ ریزی و تصمیم‌ گیری ‌های آتی سازمان قابل استفاده‌ است.نتیجه‌گیری: برای پیش بینی، درخت تصمیم، با میزان صحت 86.21% بهترین مدلی بود که در این پژوهش به آن رسیدیم و در مدل درخت تصمیم ارایه شده معیار درآمد بیمه ‌گذار به عنوان گره ریشه درنظرگرفته می‌ شود که همین نکته نشان ‌دهنده آن است روش بکار رفته می‌ تواند به شرکت ‌های بیمه کمک کند تا با تمرکز بر مشتریان سودآور به درآمد بیشتری برسند.

Discovering Profitable Customers by Data Mining Approach

Objective: Today, customers have become a critical factor in directing investors, producers, and even researchers and innovators. For this reason, organizations need to know about their customers and plan for them. Insurance companies and in general the insurance industry in each country, is one of the most important financial institutions active in financial markets, especially the capital market, which in addition to providing security for economic activities, can play a very fundamental role in providing insurance services. In other words, insurance companies play a vital role in the mobility, dynamic of financial markets and the provision of investable funds in economic activities. In this research, it has been attempted to answer one of the most important questions of insurance organizations, namely, predicting the level of customers’ losses and investing on profitable customers.Methodology: Data mining methods were used to discover knowledge to meet business needs and customer relationship management strategies. In addition, an overview of the various applications of data mining in customer relationship management in various insurance companies has been done. In the model implementation stage, a real dataset is used to evaluate the proposed model. To perform the data mining techniques in the insurance industry as data of customers, the vehicle body insurance from 2015 to 2017 has been under investigation. The total number of data used in this study from the beginning was more than 19,356, which during data preparation using Rapidminer 7.1 software became 19,356. After the initial processing, an attempt is made to extract good features from the 15 variables in the dataset that is tangible and help this research in its goal. As a result, by using clustering, drivers are divided into separate clusters based on the amount of loss, and the characteristics of each cluster are expressed. In the clustering section, three algorithms of data mining are examined. First, kmeans, kmedoids, and DBSCAN implemented on dataset. Then, the conclusion of three algorithms compared with each other based on the time of calculation and accuracy.Finding: Data mining was a good tool in this research, owing to the large volume of data, to discover the needs and identify customers. The data mining technique which was the main approach of this study fully covered the information needs by methods such as classification, prediction and clustering. The kmeans algorithm was selected as the most optimal one in time and accuracy. In the following, the implementation of the algorithms in the modeling step, the decision tree algorithm was selected and by the decision tree related to the forecasting model, it can be predicted future customers by what characteristics would be in what category. It will be valuable for the insurance companies. Using a decision tree, a forecasting model is proposed to help insurance companies to identify profitable customers which can be used for future plans of organizations.Conclusion: The customer plays an important role in today’s industry. Through studying the data obtained from customer behavior, appropriate action can be taken for marketingrelated planning and customer acquisition. The use of predictive models and preventive roadmaps has always been one of the goals of the tools that various organizations have been looking for. In this research, the insurance industry as one of the most important pillars of economic in developing countries has been chosen. By reviewing the share of the insurance industry in the economy of a developing country, it can be seen that insurance has a significant role compared to other services.  In this study, the role of insurance companies in optimizing the investment process and ways to expand the interaction between insurance examined. Customers can lead to the growth and development of the insurance industry and the capital market and thus the growth and development of the national economy. Therefore, in the implementation of this research, the data of insurance customers have been used and a forecasting model has been presented. As a good prediction model, the decision tree with 86.21% accuracy was the best model that reached in this study. The insurers’ income criterion is considered as the node root, which shows the used method can help insurance companies make more profit by focusing on profitable customers.JEL Classification: B31, C38, C22, D12