Institutional Repository of Vilnius University
Sutarties nutraukimo prognozavimas draudimo sektoriuje, naudojant mašininį mokymąsi /
Abstract
dc:descriptionThe insurance industry is often confronted with the problem of contract cancellation, where a client decides for some reason to withdraw from the services provided before the expiry date of the policy. This is due to the fact that over time the choice of policies offered by different insurance companies has been increasing, from which customers choose the most convenient and advantageous policy for themselves. For this reason, companies often have to compete with each other in order to retain customers by offering better, more individual or group-oriented offers. However, to do this successfully and to maintain profitability, more attention needs to be paid to those customers who are likely to cancel their insurance contract. This can be addressed by using machine learning techniques that predict which underwritten contracts are at risk of cancellation. Such a solution has been implemented in this study using data from one of the non-life insurance companies, which consisted of a set of attributes describing the customer, the contract concluded and the vehicle, since the focus of the study is only on policies for MTPL and CASCO products. As these situations often present a problem of class imbalance, three balancing methods were used, followed by four classification models. The results showed that balancing methods such as ENN, SMOTE or SMOTE-Tomek can be useful for obtaining more accurate forecasts of contract cancellations, and the choice of classification models is often reflected in the results. Models such as the Random Forest Classifier and XGBoost perform better when learning on data that has been balanced using the SMOTE-Tomek method, while nearest neighbour and decision tree classifiers perform better when learning on an unbalanced dataset. The best predictor of which contract will be terminated is the nearest neighbours model, which involved a feature selection step. This shows that, depending on the classifier used in a given situation, it is important to consider the importance of data balancing techniques, the selection of the most relevant features and the choice of classification accuracy metrics such as F1, recall or precision, which help to objectively assess the accuracy of the model's predictions.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Voleišo, Evelina,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD210643924&prefLang=en_US