A carregar...
Publicação
An application of predictive modeling to new business claims in automobile insurance
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.11 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
One of the biggest struggles of insurance companies is to acquire new clients and retain the most
profitable ones. Consequently, companies often accept new clients with less restricted requirements
and offer a much better price than at policy renewal. Thus, many companies make big losses in new
business. Actuary and predictive models might help to cut down this problem by discovering the
characteristics of unprofitable clients and limit the losses. This report presents a modeling approach
suggested by the company and developed during a 9-month internship.
The main objective of this project was to build the models on the probability of having a claim within
90 days from the start date of a policy in new business auto insurance. Also, the profiles of the riskiest
clients were created, and further actions defined. A comparison of the models for claims that
happened within 90 days and those that happened in the rest of the first annuity revealed some
differences in characteristics of those clients/policies. The results were obtained by training four
classifiers, namely logistic regression, decision tree, random forest, and neural network on the original
and balanced datasets. This paper discusses also different performance metrics, and an imbalanced
data problem. Finally, the paper suggests important aspects that must be taken into consideration in
future work.
Descrição
Internship Report presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics
Palavras-chave
Insurance Claims Fraud Automobile New Business Risk Predictive Models Unbalanced Dataset Logistic Regression Decision Trees Neural Networks Evaluation Metric