Tax evasion is committed in a number of ways, in which, some are easily identifiable while few others are very difficult to detect. This article deals with a sophisticated technique used by tax-evaders known as circular trading. Dealers who commit this fraud often collude together and make bogus companies using fraudulent identities with the motivation to show heavy sales transaction among them. This huge quantity of data from transactions helps the dealers to hide their actual tax manipulation. Here, we devise clustering techniques that detects and groups together the dealers who are highly susceptible in performing circular trading. We represented the entire sales database for these dealers using weighted directed graphs. The clustering algorithm is run on the commercial tax dataset provided by the state government of Telangana, India, which helped in identifying potential circular trading activities. © Springer Nature Switzerland AG 2019.