This paper investigates the issues of malicious transactions by insiders in database systems. It establishes a number of rule sets to constrain the relationship between data items and transactions. A type of graph, called Predictive Dependency Graph, has been developed to determine data flow patterns among data items. This helps in foretelling which operation of a transaction has the ability to subsequently affect a sensitive data item. In addition, the paper proposes a mechanism to monitor suspicious insiders’ activities and potential harm to the database. With the help of the Predictive DependencyGraphs, the presented model predicts and prevents potential damage caused by malicious transactions.