Desenvolvimento de modelos preditivos de mortalidade infantil com base em inteligência artificial no estado do Ceará

Abstract

The State of Ceará has registered, in the last twenty years, great progress in the planning of health care public policies that propagated the reduction in infant mortality rates, but has this reduction stopped in recent years ? Technology has also evolved in many ways recently, reducing the costs for software development, data storage and processing, the use of mobile applications that facilitate contact with population, all this allowing the large scale usage of resources such as: Data Mining, Data Mart, BI (Business Intelligence), Artificial Intelligence and others, capable of generating intelligence and knowledge to support strategic decision making for those responsable for public health planning and policies. This multidisciplinary research, involving Computer Science, Administration, Statistics and Public Health, has as its objective to develop child mortality predictive models using data mining and artificial intelligence algorithms, capable of genereting data that was not perceived or not utilized until now by public health management to support its decisions, policies and planning, aiming for the infant mortality reduction in the State of Ceará. The methodology used was the quantitative, that extracts relevant statistic information from the applied algorithms, but with bibliographical research in support of theories. This is also and applied research, for its results will be applied in the real world to benefit our community. In Ceará more than 130 thousand children are born each year and more than 1.600 die before reaching one year old, average TMI of 12,3. The main causes of infant mortality are: low weight at birth, prematurity, few or none prenatal consultation, mother’s young age and low scolarity, among others. When we apply the Neural Network algorithms thousands of times in 20 characteristic variables from 8.159 infant deaths and 10.235 living births who survived the first year of life, we reach the accuracy of 97,41% (with a margin of error of +/- 0,41%) at predicting the future. From these results we can infer that it’s possible to predict infant mortality using Artificial Intelligence algorithms with a satisfactory margin of error, allowing the utilization of this low cost technology to support public health strategies in Brazil.