Abstract:
In recent times, data mining has gained immense application because of the ability with which it can extract previously unknown and potentially useful information from raw data. Frequent Pattern Mining is a subfield of data mining in which patterns that occur frequently in the data are extracted from the data. In case of collaborative frequent pattern mining, mining may lead to the extraction of patterns that are sensitive. The revelation of such sensitive patterns is undesirable for the data owner. Privacy preservation in data mining is the area under which techniques that allow the sensitive information present in the data to be hidden from the data mining process are designed and analysed. In order to hide the sensitive information, modifications are performed on the data and this decreases the quality of the data and hence mining results obtained from such data may not be accurate. Thus, there is a trade-off between the privacy and the utility of the data. For preserving the sensitive patterns from the frequent pattern mining process various sensitive pattern hiding techniques exist. All these techniques cause side effects to the data by decreasing its quality and also are an overhead to the frequent pattern mining process. In this work the focus is to decrease the side effect caused to the data while maintaining a low running time. Existing sensitive pattern hiding techniques can be broadly categorized as heuristics based, border based and exact approaches. Heuristics based approaches are fast but they cause maximum side effect. Here we have proposed two heuristics based sensitive pattern hiding algorithms which allow fast hiding of sensitive patterns on Hadoop MapReduce framework while reducing the side effect.
