P2P lending is a new form of lending where in the lenders and borrowers can meet at a common platform like Prosper and ZOPA and strike a best deal. While the borrower looks for a lender who offers the fund at a cheaper interest rate, the lender looks for a borrower whose probability of default is nil or minimal. The peer to peer lending sites can help the lenders judge the borrower by allowing the analysis of the social structures like friendship networks and group affiliations. A particular user can be judged based on his profile and on the information extracted from his social network like borrower's friend's profile and activities (like lending, borrowing and bidding activities). We are using classification algorithm to classify good and bad borrowers, where the input attributes consists of both core credit and social network information. Most of these algorithms only take a single table as input, whereas in the real world most data are stored in multiple tables and managed by relational database systems. Transferring data from multiple tables into a single table, especially merging the social network data causes problems like high redundancy. A simple classifier performs well on real single table data but when applied in a multi-relational (Multi table) setting; its accuracy suffers from the altered statistical information of individual attributes during “join”. Therefore we are using a multi relational Bayesian classification method to predict the default probabilities of borrowers.