روش جديد و مقاوم پيش پردازش جهت بهبود تكنيك هاي داده كاوي

چكيده انگليسي :

ew and Robust Pre processing Method for Improving Data Mining Techniques Elaheh Barati e barati@ec iut ac ir Date of Submission 2011 27 04 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language FarsiSupervisors Mohammad Hossein Saraee saraee@cc iut ac ir Mohammad Reza Ahmadzadeh ahmadzadeh@cc iut ac irAbstract Over the last decades the ability of producing and collecting data has increased dramatically and thedata volume is growing rapidly These data contain valuable information These databases have becomeincreasingly large and thus more difficult to process with the available technologies The field of KnowledgeDiscovery in Databases has arisen from the need to obtain useful information from these databases and sinceits beginning it has generated a large body of research The central step in Knowledge Discovery inDatabases is Data mining which means the process of extracting implicit information which was previouslyhidden and probably will be valuable As a matter of fact Data mining aims at discovering interesting andpreviously unknown patterns form data sets The need for mining structured data has increased in the pastfew years However most data mining algorithms are not capable of working on data stored in relationaldatabases directly Most existing techniques are propositional and they extract patterns just from one table Indeed the presence of all the interesting data in a table is necessary Therefore it requires a pre processingstep for transforming relational data into algorithm specified form Unfortunately it causes to lose somevaluable information One of the multi relational data mining methods is Inductive Logic Programming ILPrequires the data to be in the form of logic clauses and it requires extra effort in preprocessing step The otherapproaches are Bayesian Networks Neural Networks Multi Relational Data Mining on relational databasesand Multi Relational Database as a Set of Trees In this research after reviewing the existing methods relational database as a set of trees method was selected By converting multi relational database into trees itis possible to apply existing tree mining techniques to identify frequent patterns in this kind of databases Thefrequent patterns that can be identified in such set of trees can be used as the basis for other multi relationaldata mining techniques In this study we proposed a new structure by combining two existing representationsfor multi relational databases which were key based tree representation and object based tree representation We used two different tree mining algorithms to identify patterns from the trees representing multi relationaldatabase based on the proposed method Moreover by applying some changes in the structure of treerepresentation we could use this structure for classification In this thesis the proposed method was appliedon a medical dataset One of the challenges of using data mining on medical data is missing values issue It isoften supposed to have a data set without missing values because if a complete dataset is used the morevaluable information will be extracted So in this study after applying different methods to deal with missingvalues the best method was selected and applied on the dataset Then by using tree mining algorithms frequent patterns were produced based on different supports These patterns were used to extract differentrules The extracted rules can provide useful information to physicians about the diagnosis Keywords Multi relational data mining relational database to trees tree mining medical data mining frequent patterns

براتي، الاهه

روش جديد و مقاوم پيش پردازش جهت بهبود تكنيك هاي داده كاوي

پايگاه داده رابطه اي به درخت , درخت كاوي , الگوهاي تكراري