Prediction of Pension Time Span in Social Security Organization Using Back Propagation Neural Networks

Social Security organization is one of the most important financial organization in Iran which is directly associated with the health and welfare of people. One of the important financial issues in this organization is the pension funds which according to the organization’s rules, in the case of death of the insured person will be paid to his survivors in order to compensate for the cutting off all or part of the income. Therefore, the main purpose of this study is to use of one of the artificial intelligence methods known as artificial neural networks to predict the duration of receiving pension based on years for the deceased’s survivors. Therefore, data related to the survivors and the main insured persons extracted from the organization’s database and after applying preprocess on them and removing the malformed data, 1498 number of data points in order to provide the artificial neural network were prepared. Then, using sensitivity analysis on neural network performance, was determined that a neural network with two hidden layers and number of neurons 30 and 18 in the first and second hidden layers, has the best performance in estimation of duration of pension receiving. By developing the mentioned neural network model in order to estimate the duration of pension receiving, based on five parameters including: the relation of the survivor with the deceased, the duration of premium paid by the deceased, gender, type of insurance for the insured person, and the insured main person s rights before death, an artificial intelligent model to predict the duration of pension receiving, was calculated. Then, the accuracy of the developed neural network, using relative error graph and regression analysis examined and revealed that the neural network has high accuracy of regression coefficient (R^2) in estimation of duration of pension receiving. In addition, accuracy of the neural network were approved by statistical parameters such as Absolute Average Relative Deviation (AARD), Standard Deviation (STD), and Root Mean Square Error (RMSE). To determine the circumstances in which the survivor will have the minimum and maximum duration of pension receiving, a series of simulation carried out on the developed neural network and using meta-heuristic particle swarm algorithm, its’ minimum and maximum poits were derived according to the results of the simulation-optimization hybrid algorithm, showed that the female survivor of an individual who have had the construction worker’s insurance, will have the shortest duration of pension receiving. It was also revealed that survivor spouse of an individual who has been insurance of elites and top talents, will have longest duration of pension receiving.

کلیدواژه ها:

Duration of Pension Receiving ، Artificial Intelligence ، Neural Networks ، Error Back propagation neural Network ، Particle Swarm Algorithm ، Social security insurance