Predictive analysis for calculating the valuation of the affiliated fund of a private pension system using machine learning techniques and tools

Abstract

This paper proposes a model for the analysis of the prediction of the accumulated fund for affiliates based on an area of study such as machine learning. The model allows to predict the pension fund of an affiliate in the private pension system by means of a web solution. In this sense, people will have, information and an adequate tool that allow them to have an oversight of the valuation of their funds throughout the years until retirement. In Peru, the decree of law 1990 states that the age for retirement is 65 years old, although there is also the option for early retirement. The proposed model consists of data analytics usage based on the modeling of machine learning algorithms through cloud platforms. The model structure includes four layers: transformation of the affiliate's data, security and privacy of personal data, obtaining and managing data, and finally, the life cycle of data applied to analytics. The model emphasizes data analytics concepts where large amounts of data are examined that lead to conclusions for better decision making. In doing this, the machine learning technique "boosted decision tree" is used due to the proximity of this technique applied in the financial forecast. The model was validated with a Pension Fund Administrator (AFP) in Lima (Peru) and the results obtained focused on the use of improved decision tree regression with a coefficient of determination of 99.997% and an average square error of 0.00650%. The coefficient of determination is an indicator of the quality of the model to predict results while the quadratic error quantifies the percentage of error among the set of results obtained under the boosted decision tree regression model.