Analysis To Urban Criminal Information Based On Statistic Algorithms

Since 1980’s, there are lots of algorithms & models merge in the scope of criminal information analyzing along with the even rapid development of computer science. We can learn many valuable quantitative researches in connection with criminal behavior property, psychological characteristics and domain distribution from references overseas. However, we find at home that there are far more qualitative researches than quantitative researches. Meantime, the study on the scope of social public order comprehensive status is still a blank field.Only if our public security department can carry out scientific analysis on criminal information and find out the implicit law within it, they can exact distinguish the "warning change" from the "routine regular change" of social public order, hence make their work progress smoothly and guarantee the orders in the political life and social life.Supported by the Chinese Security Public Ministry research project "A policing analyzing & prediction system based on data ware" (2005hbstyycx065) and "Researches on the data exchange model between policing information composite platform and police-used GIS", this paper make some researches on urban criminal information analyzing models and algorithms. In detail, major work is as follows:1. The phenomenon of "routine fluctuation" and "warning fluctuation" in field of social public order is extensively studied. On the basis of intensive analysis to present criminal information research jobs, the blank field in research scope of tendency analysis is put forward. Afterwards, the feasibility of applying intrusion detection idea into criminal information analysis is discussed, some problems which need to carefully deal with are also raised.2. A profile vector is proposed to represent the status of social public order. This vector is formed by rotate and superposed several measure distributions so that it can represent the current criminal statistics property as well as the historic one. Since the profile vector repeatedly considers the effect of historic factor, it can better represent the time-fluctuated property of the urban crime statistics. A Chi-square testing based Distribution Classifying Algorithm (CDCA) is proposed to pre-process the raw data of criminal statistics. CDCA classifies the raw data in order to obtain certain kind of data classes, and then Chi-square testing is used to verify the Poisson distribution of these classes. Considering the testing interval is single side sliding, CDCA can obtain the optimized interval of Poisson approximation.3. A Composite Statistic Profiling-Vector Hypothesis-test Algorithm CSPHA is presented to testing the system exception index. CSPHA design three kinds of time span to respectively collect data, abstract data, analyze data from scattered data centre. After profile vector was generated, a system exception testing expression and a system exception referenced expression defined in CSPHA are used to exam the present statistics value set. Experimental system and experiment results show its precision.4. An Enhanced Composite Statistic Profiling-Vector Hypothesis-test Algorithm (ECSPHA) is proposed to test the present statistics data in combination with the interrelation between different profile vector components. ECSPHA select covariance of statistics data as a substitute for covariance of profile vector in order to amplify the interrelation property between different crimes. The exception testing measure is optimized in order to improve its sensitivity to the interrelationship between different measures of complex system. Experiments show that ECSPHA is not sensitive to single statistic data; on the contrary, it shows sensitivity to relative statistics data.5. A Long-term Historical Profile based Information Processing Algorithm (LHPIPA) is proposed. LHPIPA computes the historic distribution characteristic of every present statistic data, and then normalize these data with a standard normal distribution. LHPIPA sufficiently consider the expert opinion and introduced it to intervene in decision-making by design an exception testing expression including a series expert index. Experiments show that LHPIPA possess better expert decision property and a relatively long term testing span.