Mining malicious corruption of data with Hidden Markov models

Data mining algorithms have been applied to investigate a wide range of research issues recently. In this paper we describe an alternative technique of profiling databases via time series analysis to detect anomalous changes to a database. We view the history of modifications in the database as a set of time series sequences. We then examine the application of Hidden Markov models (HMMs)as a mining tool to capture the normal trend of a database's changes in transactions. Rather than examining each record independently, our technique accounts for the existence of relations among groups of records, and validates modifications to the sequence of transactions. The algorithm is adaptive to changes in behavior. Experiments with real data-sets, comparing various options for the initial HMM parameters, demonstrate that the distribution of the change in acceptance probabilities of anomalous values is significantly different from that of acceptance of transactions expected by the model.