Publicação
Enhancing Fraud Detection in the Insurance Industry: Integrating SHAP explanations into an Unsupervised Anomaly Detection Model with Autoencoder-based Dimensionality Reduction
| datacite.subject.fos | Ciências Naturais::Ciências da Computação e da Informação | pt_PT |
| dc.contributor.advisor | Henriques, Roberto André Pereira | |
| dc.contributor.author | Costa, João Filipe Dias Gomes de Morais | |
| dc.date.accessioned | 2024-03-07T18:48:10Z | |
| dc.date.available | 2024-03-07T18:48:10Z | |
| dc.date.issued | 2024-02-02 | |
| dc.description | Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Data Science | pt_PT |
| dc.description.abstract | In the insurance sector, machine learning techniques are widely employed to aid auditing teams in identifying potentially fraudulent claims. At Future Healthcare Group, an unsupervised anomaly detection (UAD) model has been deployed to support a dedicated team in the audit process. This model incorporates an autoencoder for dimensionality reduction of part of its feature space. This project starts with the question: 'Is it possible to increase the efficiency of the current UAD model by increasing its interpretability with SHapley Addictive Explanations (SHAP)?'. Due to its 'nested architecture' the direct implementation of SHAP explanations directly into this model poses computational challenges namely in uncovering the information compressed by the autoencoder. This project aimed at developing a framework that efficiently integrates SHAP explanations into the unsupervised anomaly detection model. This project is divided in two steps: In the first step, it focuses on building the framework; in the second, the framework output is evaluated. The framework increased the efficiency of the model. This was achieved mainly by indirectly increasing the UAD model performance. The presence of the explanations allowed to uncover observations classified as anomalous due to its rarity that were not true anomalies by business definition. This allowed the pre-filtration of these, which contributed indirectly to the increased performance of the based model. In summary, the developed framework offers an efficient solution for integrating SHAP explanations into an unsupervised anomaly detection model, particularly when a part of the feature space undergoes compression via an autoencoder. | pt_PT |
| dc.identifier.tid | 203544439 | pt_PT |
| dc.identifier.uri | http://hdl.handle.net/10362/164589 | |
| dc.language.iso | eng | pt_PT |
| dc.rights.uri | http://creativecommons.org/licenses/by/4.0/ | pt_PT |
| dc.subject | Health Insurance | pt_PT |
| dc.subject | Fraud Detection | pt_PT |
| dc.subject | Unsupervised Learning | pt_PT |
| dc.subject | Anomaly Detection | pt_PT |
| dc.subject | Autoencoder | pt_PT |
| dc.subject | SHAP | pt_PT |
| dc.subject | SDG 8 - Decent work and economic growth | pt_PT |
| dc.subject | SDG 9 - Industry, innovation and infrastructure | pt_PT |
| dc.subject | SDG 16 - Peace, justice and strong institutions | pt_PT |
| dc.title | Enhancing Fraud Detection in the Insurance Industry: Integrating SHAP explanations into an Unsupervised Anomaly Detection Model with Autoencoder-based Dimensionality Reduction | pt_PT |
| dc.type | master thesis | |
| dspace.entity.type | Publication | |
| rcaap.rights | openAccess | pt_PT |
| rcaap.type | masterThesis | pt_PT |
| thesis.degree.name | Mestrado em Ciência de Dados e Métodos Analíticos Avançados, especialização em Ciência de Dados | pt_PT |