Identifying Suspicious Records in Household Data

Abstract

With growing advancement in the financial field, fraud is spreading all over the world, causing major financial losses. This thesis represents methods to detect suspicious records in household income and expenditure survey gathered by the census and statistics department in Sri Lanka. K-means, hierarchical clustering method and model base clustering method were used in this research to find the suspicious records in the in household income and expenditure dataset. Some statistical methods were also used to clarify the suspicious records. In the dataset when the adhoc expenses such as weddings /funerals for family members, Social activities / Ceremonies, gift, donation and purchased properties house exceed the limit, income is higher than the expenditure without proof those are considered as a suspicious record.