Resumo

GOMEZ-JARAMILLO, Paola Andrea; GONZALEZ-ECHAVARRIA, Favián  e  PEREZ-RAVE, Jorge Iván. Technological incident classification model from a machine learning approach in insurance services. Dyna rev.fac.nac.minas [online]. 2022, vol.89, n.221, pp.161-167.  Epub 14-Set-2022. ISSN 0012-7353.  https://doi.org/10.15446/dyna.v89n221.100070.

Managing technological incidents in insurance companies requires a correct and timely assignment of these to the problem-solving teams. Classifying these incidents by humans demands time and knowledge and is frequently executed erroneously. This paper addresses this classification problem from a machine learning approach. The performance of five supervised learning methods (logistic regression, classification trees, random forest, discriminant linear analysis, support vector machines) is compared in three scenarios of inclusion of predictors: structured, texts, and both. The use of unstructured variables considerably improves the accuracy of the models (e.g., Random Forest, validation sample: 0.709 using structured data; 0.881 using text data). Moreover, considering the practical implications of the human correct classification rate (66%) vs. machine (88%, Random Forest, SVM, or linear regression), the machine favors resource-saving in the organization. This article is a successful case of machine learning in the insurance industry.

Palavras-chave : classification of incidents; technology incidents; insurance; machine learning.

        · resumo em Espanhol     · texto em Inglês     · Inglês ( pdf )