ارائه ي يك روش خودكاربه منظوراستخراج كنش هاي مقرون به صرفه از داده ها با استفاده از تئوري فازي

چكيده انگليسي :

An Atomatic Method for Discovering Cost Effective Actions from Data by Using Fuzzy Set Theory Nasrin Kalanat n kalanat@ec iut ac ir Date of Submission 2012 02 27 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language FarsiSupervisor Mohammad hosein Saraee m saraee@cc iut ac irAbstract Data mining is the process of discovering valid novel and understandable patterns from datawhile the discovered patterns can be usable and actionable for business decisions most ofmachine learning and data mining techniques only focus on finding frequent patterns and usuallydo not pay any attention to actionably and usability of mined patterns Thus data mining isconverted to a data driven trial and error process that faces users with many patterns and they willbe confused about how and what to do with them It is because the discovered knowledge by datamining methods is not actionable to satisfy the real world requirements In order to solve thisproblem data mining must be developed towards real world business For this developing it needsto consider domain factors and constraints in data mining process Actionable KnowledgeDiscovery is a paradigm shift from data driven data mining toward domain driven data miningthat is aimed at discovering actionable knowledge to satisfy real world requirements Up to now many researches have been done on AKD that have considered some of real worldfactors and constraints such as changeable or not changeable attributes cost of change of value sattributes distributed data incomplete data All of these researches assume that the data is precise while in most of real world scenarios we never face with quite precise values and always face witha degree of uncertainty Therefore accuracy or precision in real world situations is simplifying andidealizing that cause high rate of error in AKD methods Consequently some useful actionableknowledge strategies can be missed during the search process of these methods and even somenon actionable knowledge strategies may be produced In this thesis these drawbacks will beovercome by fuzzy set theory We propose the idea of combining fuzzy set theory with actionableknowledge In this regard and in order to improve the only existent method on discovering actions Leaf Node Search Method one type of fuzzy actionable knowledge named fuzzy action will beintroduced and the idea of considering cost of change of attribute values as a function will beproposed Also a profit function will be presented for predicting the net profit of each fuzzy action In this thesis we present an algorithm that suggests fuzzy action in order to decrease thedegree to which a certain object for example a customer belongs to an undesired status andincrease the degree to which it belongs to a desired one while maximizing objective function namely the expected net profit The contribution of this thesis is in taking the output from fuzzydecision trees for producing actionable knowledge through automatic post processing The effectiveness of our proposed method will be verified on four public data set of UCI incomparison with the leaf node search method It will be shown that our method is more efficientthan Leaf Node Search method in a number of suggested actions total profit and average profit Keywords Actionable knowledge Fuzzy action Fuzzy decision tree Post process

كلنات، نسرين

ارائه ي يك روش خودكاربه منظوراستخراج كنش هاي مقرون به صرفه از داده ها با استفاده از تئوري فازي

كنش فازي , درخت تصميم گيري فازي , پس پردازش , دانش قبول كاربرد