Predictability of loyalty and separation of self-insurance Persons of Social Security Organization based on data mining method

Document Type : Original Article

Author

Social Security Organization

Abstract

Loyalty and customer loyalty is one of the major challenges of companies and institutions that provide services and goods to their customers. The social security organization has faced the challenge in recent years in the issue of loyalty of its insured individuals to continue insurance. In this research, information and important indicators of 21407 people were extracted from 27 databases of this organization. Then, using the NSGA-II algorithm, seven important characteristics with the least degree of classification error were selected. In the first step, the classification of data with multi-layer neural network was performed on existing data using 27 characteristics and the classification accuracy was 97.6%. After selection of the best class that was neural network, classification operation with neural network was performed on the same data and in the form of 7 characteristics and accuracy of 96.8%. At the selection stage, the best stratified strain was used to perform the required predictions from three multi-layer neural network algorithms, backup vector machine (SVM) and KNN algorithm. Finally, the multilayer neural network had the best degree of accuracy with a 96.8% got. Then, for the neural network, the training was carried out using 7 characteristics related to the data of years 1367 to 1395. The trained neural network was used to predict loyalty and customer rejection of 1396 and 1397, with 8364 records. Finally, it was concluded that considering all probabilities, about 27.65% of the insured persons who were born in 1396 and 1397 would be categorized in a reversal class.

Keywords