Life insurance is an important branch of insurance industry, which has huge market development space. Therefore, with the opening of the life insurance market and the intervention of foreign companies, the competition has gradually escalated, and the competition between competitors is a foregone conclusion. How to maintain the core competitiveness and make yourself invincible is a problem that every enterprise must face. The application of information technology is undoubtedly one of the effective means to improve the competitiveness of enterprises. After years of development, the life insurance information system has gradually matured and improved, and accumulated a considerable amount of data resources, providing a solid foundation for data mining. However, life insurance companies pay more and more attention to discovering knowledge through data mining and using it for scientific decision-making.

data mining

Data mining refers to the process of extracting useful information and knowledge from a large number of incomplete, noisy, fuzzy and random data. Its manifestations are concepts, rules and patterns.

At present, there are many mature data mining methods in the industry, which provide an ideal guiding model for practical application. Crisp-DM (standard process of cross-industry data mining) is one of the recognized and influential methods. CRISP-DM emphasizes that DM is not only data organization or presentation, but also data analysis and statistical modeling, and it is also a complete process from understanding business requirements, seeking solutions to being tested by practice. CRISP-DM divides the whole mining process into the following six stages: business understanding, data understanding, data preparation, modeling, evaluation and deployment.

Business understanding is the understanding of enterprise operation, business process and industry background; Data understanding is the understanding of existing enterprise application systems; Data preparation is to extract a subset of sample data related to the problem to be explored from a large number of enterprise data. Modeling is to choose a more practical mining model and form a mining conclusion on the basis of understanding business problems and preparing data. Evaluation is to test the conclusion of mining in practice, and if the expected effect is achieved, the conclusion can be published. In practical projects, data understanding, data preparation, modeling and evaluation in CRISP-DM model are not one-way operations, but a process of repeated, adjustment and constant correction.

Industry data mining

After years of system operation, life insurance companies have accumulated quite a lot of policy information, customer information, transaction information, financial information and so on. And there is a very large-scale database system. At the same time, data concentration provides conditions for upgrading the original business level and expanding new business, and also provides rich soil for data mining.

According to CRISP-DM model, the first thing data mining should do is to understand the business and find the goals and problems of data mining. These problems include: agent selection, fraud identification and market segmentation, among which market segmentation has important guiding significance for enterprises to formulate business strategies, and is the primary issue related to the survival and development of enterprises and the formulation and realization of enterprise marketing strategies.

According to the characteristics of life insurance management, customer groups can be classified and summarized from different angles, thus forming various customer distribution statistics as the basis for managers to make decisions. Starting with life insurance products, it is relatively easy to analyze customers' preferences for different types of insurance and guide agents to focus on promotion. Because the domestic economic development is different and there are great differences among provinces, it is necessary to limit the sampling of analysis data to an area with the same economic level. At the same time, market fluctuation is also a problem that must be considered. A model has a life cycle from establishment to abandonment, which is determined by the adaptability and hit rate of the model, so the model needs constant revision.

Mining system architecture

Mining system includes two parts: rule generation subsystem and application evaluation subsystem.

The rule generation subsystem mainly completes statistics, generates relevant rules and outputs relevant results according to the historical data of insurance policies provided by the data warehouse. Specifically, it includes data extraction and transformation, mining database establishment, modeling (including parameter setting), model evaluation and result release. The target of publication is high-level decision makers, and the model is submitted to the application evaluation subsystem, and a new model is dynamically generated every month according to the effect.

The application evaluation subsystem can be understood as a mining agent in the production system, which makes a classless prediction of the strategy data according to the rules and certain strategies generated by the generation subsystem. Through systematic task planning, the evaluation index of production data is generated. Specifically, it includes automatic transfer of core business system data to data platform, real-time evaluation of rules, dynamic display of evaluation results and actual effect evaluation. The rule evaluation subsystem detects according to the rules. After a period of testing, the rule generation subsystem can be used to relearn, obtain new rules and update the rule base continuously until the rule base is stable.

At present, the commonly used analysis indicators are: insurance, payment period, insured occupation, insured annual income, insured age, insured gender, insured marital status and so on.

In practice, we can choose factors appropriately according to the actual data, and make different degrees of summary, thus forming a satisfactory decision tree and producing interpretable conclusions.