# Introduction loud computing is derived from technologies such as distributed processing, parallel processing, grid computing, etc. It is an emerging approach to sharing the infrastructure architecture [1]. It distributes all the computing tasks on the resource pool that is made of many computers, making sure all the application systems can acquire desired computing power, memory space and software service according to their demand [2]. All the computing is provided to the terminal user by the form of service, and all the application software in the cloud as shared resources. A cloud database is a database deployed and virtualized in the cloud computing environment. It is predicated that as it develops overtime, more and more people and companies will store all their data in the cloud, which will make data mining based on the cloud computing one of the trends in the future data mining systems [3]. There is a massive amount of data in the cloud database, and among them, lives potentially valuable knowledge. How to discover such useful knowledge is the key point in database research. Data mining is the process of picking out the hidden knowledge and regulations, which possess potential value that could influence decision making [4]. Data mining namely refers to the knowledge discovery from a database and is comprised of the following procedures: data preprocessing, data alternating, data mining operation, rule expression and evaluation [5]. A data mining system includes: control unit -used to control all parts in a harmonious way; database interface -used to generate and process data according to the given query; database -used to store and manage relevant knowledge; focus -refers to the data extent that needs to be inquired; model extracting -refers to the various data mining algorithms; and finally, knowledge evaluation-used to evaluate the extracted conclusion [6]. # II. # Cloud Database A distributed database is a logical set of the databases at various sites or nodes in a computer network and logically, such databases belong to the same system [7]. Different from the traditional distributed database, a cloud database contains isolated as well as shared data; a cloud database can be designed by using different data models, which mainly include the key-value model and relationship model. All data of the key-value model, including the rows and columns, are stored in the cells of a table. Contents are partitioned by row, the rows make up a tablet, and the tablet is stored on a server node. # a) Row Key Data is maintained in the lexicographic order on the row key. For a table, a row interval is dynamically partitioned according to the value of the row key and is the basic unit in which load balancing and data distribution are performed. Row keys are distributed amongst data servers. # b) Column Key Column keys are grouped into sets of many "column families" and are the basic units in which access control is performed. All data stored in a column family usually belong to the same data type, which means data is compressed at a higher rate. Data can be stored in a column key of the column family. # c) Timestamp Each cell contains multiple versions of the same data and these versions are indexed by the timestamp. Data model for key-value cloud database is as shown in Fig. 1: Figure 1 : Data model for key-value cloud database "com.cnn.www" "contents" "anchor:cnnsi.com" "anchor:mylook.ca" "" "" "" "cnn" "cnn.com" t5 t9 t3 t6 t8 The data model for the relational cloud database involves such relevant terms as row group and table group. A table is a logical relationship and includes a partitioning key, which is used for partitioning the table. The set of many tables with the same partitioning key is called a table group. In that table group, the set of rows with the same partitioning key value is called a row group. The rows in that row group are always allocated to the same data node. Each table group contains many row groups, which are allocated to different data nodes. A data partition contains many row groups, so each data node stores all rows with a certain partitioning key value. The data model for the relational cloud database is as shown in Fig. 2: The normal target of the association rules is to discover the data relations among the data item set in the relationship type cloud database. Through mining based on the association rules, we can discover the relevance of the data. In the subject item set, there are some target features in the relationship type cloud database. For instance, the commodity data item set in the commercial behavior analysis {T-shirt, coat, shoes, milk, bread ... }; data item set in the medical diagnosis analysis {hypertension, diabetes... }. Classifying item set has the similar features with the subject item set, for instance, customer data item set in the commercial behavior analysis {vocation, gender, age... }; diagnosis behavior in medical diagnosis and signs and symptoms item set {smoking, polysaccharide, hyperlipidemia ... }. Sample item set, which has both the features in the subject item set and the transaction data item set in the classifying item set. For instance, transaction data in the commercial activity analysis { {Zhangsan, milk}, {Zhangsan, bread}, {Lisi, T-shirt} ... }, health check information in the medical diagnosis {{ Zhangsan, smoking, hypertension}, {Lisi, hyperlipidemia, diabetes}... }. Through the mining based on the association rules, we can find that 90% of the customers who bought milk also bought bread; 50% of the patients who have hyperlipidemia also have diabetes. The common targets of the association rules are transaction databases with the characters of subjects oriented item set. In practice, most databases are relational, and many applications and the required knowledge are from many different item sets(or multiitem set for simplicity). For relational databases, it is difficult to describe the complicated association rules between the multi-item set with models of general association rules. We present the association rules model of multi-item set for the relational databases: Definition 1: I is the subject item set, J is the taxonomy item set, each transaction corresponds to a subset T of the subject item sets and a taxonomy item U of the taxonomy item sets, called T belonging to class U. Model 1: it is supposed that R=(r1,r2,?, rn) is the rows group in the relational cloud database, rk is one of the rows item set, D is a sample item set relevant to R, and each sample d corresponds to one rows item set, i.e. d?R. Each sample is marked with SID (sample identifier). As for the classifying item set X, only when X?d, the sample X belongs to d. association rules is a formula like X?d?Y?d, it can be X?Y, therein, X?R, Y?R and X?Y=?. The rule X?Y in the sample item set D is constrained by degree of confidence C and degree of support S. Degree of confidence C is defined as C% in the transaction X in D also contains Y. Degree of support S is defined as transaction X?Y accounts for S% in D. Degree of confidence represents the strength of the rule, while Degree of support means the frequency of the model, which is shown in the rule. In the cloud database containing cases information, 66% of the crime site in the theft cases happened in factories, so the C is 66%. Theft cases and factory cases account for 17% of the total cases, so the S is 17%. The data frequency item set can be defined as the data item set where the degree of support S is over the pre-defined minimum degree of support S. The association rules with high degree of support S and degree of confidence C is considered strong association rules, otherwise it is considered weak association rules. Association rules mining means to find the line group that accord to the strong association rules in the database. The procedure for mining these kinds of association rules of multi-item set is as follows: 1. Divide transaction D into several transaction subset D'={D1',D2',?Dn'} according to taxonomy item sets. # For all D1'=Cp uj?{u-u relates to ui } End If Nexts L=?{u-u relates to ui } Next Output L Following the idea of converging classes, the case information is divided into many kinds with the equivalent dividing method. Define the base value of kinds as classifying the support degrees. The kinds can be divided into strong class ones and weak class ones. The weak class has too small classifying support degrees, no practical meanings and can be neglected. For the strong class, Rough set theory can be used to analyze their common features and form the classifying characteristic regulations. Data was mined from the database for the experimental criminal case information system by using the aforementioned algorithm. Taking the crime approach table group as an example, the table contains 100762 records. Given a classification support degree threshold of 10%, and a characteristic confidence degree threshold of 20%, 359 classification characteristic rules were mined, for example: (Residential house, night, 23.4%) (Rubbery, less than RMB10000, 93.3%) VI. # Conclusions The data mining technique is new to the information society. Many subjects need to be studied in this field. In many professions, a certain amount of databases have been accumulated, in which some hidden knowledge needs to be discovered. Starting with the concept of set theory, the data model for the cloud database was analyzed; the model and algorithm for mining classification characteristic rules from cloud database were designed to make data mining of classification characteristic rule more practical. The abstracted related knowledge models presented in this paper can be put into practice in the public security affairs, such as case chaining, which is one of the highly demanded, complex tasks in the public security affairs. The presented methods about the related case data mining in this paper promote the work effect of the chained cases. On the case material analysis, the mining of the classifying characteristic regulations help users with their classifying work and overcome the weaknesses that exist in the old statistics method, in which repeated experimentation are required. 2![Figure 2 : Data model for relational cloud database III. Data Mining for Association Rules a) Model of the Association Rules](image-2.png "Figure 2 :") for(j=1;j<=m;j++) dobeginL j,1 ={large 1-items};for (k=2;L j,k-1 ??;k++) dobeginC k =apriori-gen(L j,k-1 );forall samples s?D j dobeginCs=subset(C k ,s);forall candidates c?Cs doc.count++;endL j,k ={c?C k |c.count>=minsup}endAnswer=? j,k L j,k ;end;Lj,1 1AA1A2A3k 1v 11v 20v 32k 2v 10v 21v 30k 3v 12v 20v 30k 4v 11v 21v 30k 5v 11v 20v 32k 6v 12v 20v 30k 7v 10v 21v 31k 8v 11v 21v 31k 9v 11v 20v 32k 10v 10v 21v 31k 11v 11v 20v 32k 12v 10v 21v 31 E?T?Cp)E: classT: characteristicCp: confidence degree of characteristicDiscoveryAlgorithmforClassificationCharacteristic Rules.D is a cloud database and A is the set of classificationattributes of D.For all A t For j=1 To k Doif C i ?C p Then(E i , T j , C j ) ?result baseEndifNextEndifNextNextV. 2Case kindSelected siteWay of commitCase1Burglaryresidencedoor pickedCase2Burglaryresidencedoor smashed © 2014 Global Journals Inc. (US) © 2014 Global Journals Inc. (US) Global Journal of Computer Science and Technology * Research on cloud databases ZYLin YXLai CLin YXie QZou Journal of Software 23 5 2012 * Proc. of the 1st Int'l Conf. on Cloud Computing MGJaatun GSZhao CMRong of the 1st Int'l Conf. on Cloud ComputingBerlin Springer-Verlag 2009. 2009 * Study on cloud computing security DGFeng MZhang YZhang ZXu Journal of Software 22 1 2011 * Application of the data mining technique in case information systems TXZhu W PZhang D ICCSEE 2012 1 * Technology for Mining Classification-Characteristic Rules TXZhu LLi ZWXu Journal of Shenyang Polytechnic University 1999 * Cloud Computing-Oriented Data Mining System Architecture JJHe CMYe XBWang ZXHuang QLLiu Application Researche of Computers 28 4 2011 * HadoopDB: An architectural hybrid of MapReduce and DBMS technologies for analytical workloads AAbouzeid KBajda-Pawlikowski DJAbadi ARasin ASilberschatz PVLDB 2 1 2009