Journal of Quality Engineering and Management

Abstract Recent years, the insurance industry has been experiencing an increase in equipping insurance companies with fraud detection systems. Furthermore due to the significant cost imposed on the insurance industry by the rise in such claims, the role of data mining techniques in detecting fraudulent claims has become widespread. However an essential issue with such systems is the quality of their outputs. On one hand, supervised algorithms are more accurate comparing to unsupervised counterparts. On the other hand, as data labeled fraud is really limited, the efficiency of supervised algorithms is severely challenged. Within this regard, a novel approach is introduced as “alternative feature” to overcome the challenge. Basically, alternative feature is a variable whose values are available and can be considered a suitable indicator to detect suspicious cases. This approach improves the efficiency of the system and allows experts and insurance companies to investigate suspicious cases with more confidence and less error.