KSEF: Building a Preliminary Prototype Terrorist Analysis System with Privacy Protection

Detecting terrorist groups from vast population without too many false positives is a challenging task for computer scientists. The sheer amounts of data and the high dimensionality of attributes to be analyzed make such a task algorithmically difficult and computationally intensive. An important aspect of terrorist detection models is the ability to detect small scale local correlations against a background of large scale diffuse correlations. Newly developed data mining techniques based on matrix decompositions look into subclusters within larger clusters, and transform correlation into other properties that may be distinguished more easily. Widely used spectral analysis techniques in data mining, such as the singular value decomposition (SVD), can transform correlation relationships into regions of increased proximity in lower dimensions. Privacy is one ofthe major concerns in many data mining techniques. We will construct a preliminary prototype terrorist analysis system with privacy protection, by taking advantage of SVD. Only the data owners and the authorized users can access to the original data. The analysis is done on the transformed datasets. Taking advantages of our experience and expertise in data mining, information retrieval, and matrix computation, we will develop application specific data mining techniques and software that can be used to detect local correlations. Due to the inherent difficulty and complexity of this work, and the sensitivity of data privacy, following the standard practice in this domain of research, we will generate artificial datasets to test our techniques and software. This work is an effort to meet the President's definition of homeland security, i.e., a concerted effort to prevent terrorist attack within the United States, and to reduce America's vulnerability to terrorism.