NORMA eResearch @NCI Library

Detecting Anomalous Insurance Claims with Hybrid Feature Optimisation and Classification Techniques

Dasgupta, Sananda (2019) Detecting Anomalous Insurance Claims with Hybrid Feature Optimisation and Classification Techniques. Masters thesis, Dublin, National College of Ireland.

[thumbnail of Master of Science]
PDF (Master of Science)
Download (5MB) | Preview
[thumbnail of Configuration manual]
PDF (Configuration manual)
Download (2MB) | Preview


As the world is gradually being engulfed by the inevitability of technology, each and every aspect that technology is incorporated into are becoming increasingly vulnerable to digital crimes. The insurance sector is no exception to this and the potentiality of insurance frauds are taking a huge toll on the industry and the numbers are increasing day by day. Basically, any act carried out to swindle an insurance process can be termed as an ‘Insurance Fraud’. Fake claims are another way by which a malefactor can deceive an insurance process. Such is the magnitude of the threat that the industry loses almost $30 billion a year according to a recent survey. Several methods and processes have been applied and tested as an anti-fraud measure to minimise and ideally terminate illicit activities in the sector and data-mining methodologies have proven to be instrumental in
fighting digital crimes in the insurance domain. Although there exists several ways and methods of applying data-mining into a fraud-prevention program, this research
particularly aims to explore an optimal hybrid model in identification of aberrant and atypical activities in an insurance claim process in an attempt to detect potential
anomaly. The efficiency of this particular model that combines feature optimisation with classification algorithms is based on the performance metrices viz. accuracy, sensitivity and specificity. The model is being tested on a dataset of insurance claim taken from Kaggle and the feature optimisation algorithms used are Particle Swarm Optimisation (PSO) and Firefly Algorithm (FFA). The classification algorithms applied are Support Vector Machines (SVM), Artificial Neural Network (ANN), Naïve Bayes (NB), KNearest Neighbour (kNN) and Random Forest (RF). In an attempt to achieve a high quality predictive output on the basis of the above-mentioned metrices, this paper investigates that a hybrid combination of PSO and RF proves to be the most effective in achieving the best predictions over other models. This research is optimistic that the model deduced will hence allow insurers to put a check on fraudulent activities within the industry that eventually will save billions of dollars globally.
Keywords: Data Mining, Feature Selection, Fraudulent Insurance Claims, Particle Swarm Optimisation (PSO), Firefly Algorithm (FFA), Classification Algorithm,
Artificial Neural Network (ANN), Support Vector Machine (SVM), Random Forest (RF), Naïve Bayes (NB), K-Nearest Neighbour (kNN).

Item Type: Thesis (Masters)
Subjects: Q Science > QA Mathematics > Electronic computers. Computer science
T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science
Q Science > QA Mathematics > Computer software
T Technology > T Technology (General) > Information Technology > Computer software
H Social Sciences > HG Finance > Insurance
Divisions: School of Computing > Master of Science in FinTech
Depositing User: Dan English
Date Deposited: 03 Jun 2020 09:33
Last Modified: 03 Jun 2020 09:33

Actions (login required)

View Item View Item