Author Email: firstname.lastname@example.org
The fraudulent activities are increasing day by day with increase in technology in insurance sector. These fraud cases make shoddy impact on socio-economical system. This paper presents a detail survey of machine learning techniques used in insurance fraud prediction. This paper has disclosed traditional machine learning techniques like supervised and unsupervised learning and also some contemporary methods such as hybrid and ensemble leaning. The approach of the problem changes with the change in dataset hence this paper aims to provide an organized overview of the fraud prediction techniques based on the type of training data provided to the machine learning model. Keywords Insurance Fraud; Supervised Learning; Unsupervised Learning; Hybrid Classifiers; Ensemble Classifiers; Bagging; Boosting; Stacking.
‘Insurer’ and ‘Insured’ are two pillars of insurance industry and the whole business runs due to the utmost faith within both of them. Insurance Fraud occurs when any action performed by either insurer or insured with an aspiration to gain some advantage to which they are not legally permitted or fraud may occur when any one of the party purposely refuses to provide benefits to other party which was legally permitted to them. The main intention behind initiating an insurance fraud is “to appear as conventional and to be proceed and get recompense in routine manner” The survey of TOI reveals that one in every ten insurance claims is found to be fraud, which means around 10% of total insurance claims are fraud. According to the study of KPMG India Financial Services, insurance is the most vulnerable to fraud than other financial services, the survey says that loss caused due to insurance frauds are over Rs. 30,000crores(approximately $45billion) which is actually 9% of the total amount of insurance industry. The financial survey of Ernst & Young says that premium, claims and third party frauds are the three main fraud risks in insurance sector, from which only fraudulent claims contributes around 50% of the total fraud.
This survey has explored the machine learning techniques used in insurance fraud prediction. Machine learning approach provides the vast range of methods and algorithms for fraud prediction. Supervised and unsupervised learning methods are widely used in combination with other methods to improve the prediction accuracy of the model. Hybrid learning methods provides flexibility to user by blending different algorithms together these techniques have outperformed than that of the traditional learning methods. Ensemble learning is gaining more importance recently due to its reliability and flexibility with different approaches. In few recent studies it is revealed that ensembles not only improve prediction accuracy but they also deal with some chronic machine learning problems such as over-fitting, class imbalance and concept drift. Ensemble models and their applications are tempting because of their generalization ability. Ensembles are expensive to build in terms of both time and resources but this could be seen as one time investment because once the ensemble is assembled it produces highly efficient results.
507 total views, 1 views today