Case Studies On Data Mining In Market Analysis

International Journal of Engineering Trends and Technology (IJETT) – Special Issue – April 2017Case Studies On Data Mining In MarketAnalysisShravan Kumar Manthri 1 and Hari Priyanka Chilakalapudi 2Department of Computer Science and Engineering1Bhoj Reddy Engineering College for Women,Hyderabad-500059, Telangana,IndiaDepartment of Computer Science and Engineering2Bhoj Reddy Engineering College for Women,Hyderabad-500059, Telangana,IndiaAbstract- A huge chunk of data is generated each minutein enterprise business. Extracting information from pilesof data helps in extracting patterns that can predict andguide future behaviour of the enterprise. Data miningtechniques filter through large amounts of raw data andextract useful information that gives enterprisebusinesses a competitive edge in the market. Variouscases on customer purchasing habits have beenpresented and also used in real problems. Data miningtechniques are highly effective in analysing consumerbehaviours. It helps enterprises to make informedbusiness decisions, enhances business intelligence,thereby improving the company’s revenue, detectinganomalies, fraud detection and reducing cost overheads.In this paper the author reviews various such techniques.Further, the application of these techniques in variousscenarios is analyzed.I.IntroductionData miningData mining is a process used by companies toturn raw data into useful information. By usingsoftware to look for patterns in large batchesof data, businesses can learn more about theircustomers and develop more effectivemarketing strategies as well as increase salesand decrease costs. Data mining depends oneffective data collection and warehousing aswell as computer processing. To study acustomerpsychologicalmindsetandconverting this into statistical format and seethat if there is any technical format by whichwe can analyze his buying ry stores are well-known users of datamining techniques. Many supermarkets offerfree loyalty cards to customers that give themaccess to reduced prices not available to nonmembers. The cards make it easy for stores toISSN: 2231-5381track who is buying what, when they arebuying it and at what price. The stores canthen use this data, after analyzing it, formultiple purposes, such as offering customerscoupons targeted to their buying habits anddeciding when to put items on sale or when tosell them at full price. Data mining can be acause for concern when only selectedinformation, which is not representative of theoverall sample group, is used to prove acertain hypothesis.Power of hidden information in data:Enterprises generate terabytes of data each daythat is stored in databases, data warehouses, orvarious other kinds of data repositories. Mostof the valuable information may be hidingbehind such data; the overwhelming datavolume makes it difficult for human beings toextract them without the help of powerful toolsand techniques. At the beginning of the lastdecade information was only available onpapers and only at a specific time. In this age,information is now easily accessible, ascontent providers, content locators andpowerful search engines have enabled accessto huge amounts of information in no time.Natural Language Processing has allowedaccess to the same information in a languagefamiliar to the person. From the time whenpeople used to wait in huge queues to paybills, taxes or for movie tickets andentertainment, we have reached a time whereeverything happens in a few clicks. All thesechanging factors and trends made a hugeimpact on the enterprises which are highlydependent on the consumer behaviour andbasket trends. Every enterprise needs to behighly scalable and be able to predict thefuture behaviour of the customer to makeprofits in their business. This very needs leadto the Market Analysis. Earlier techniques likequestioners, surveys, test marketing, previoushttp://www.ijettjournal.orgPage 180

International Journal of Engineering Trends and Technology (IJETT) – Special Issue – April 2017sales data and leading indicators were used topredict the future of a product. All thesetechniques although proved to be effective,were highly time consuming and required a lotof manpower. Data mining techniques havecompletely changed the scenario. Informationwhich was once gathered by travelling miles isnow available in few seconds.II. Data Mining TasksClassification:Classification is the process of finding a modelthat describes the data classes or concepts. Thepurpose is to be able to use this model topredict the class of objects whose class label isunknown. This derived model is based on theanalysis of sets of training data.Eg: 1.Assigning voters into known buckets bypolitical parties .2.Bucketing new customers into one ofknown customer groups.Regression:In statistical modeling, regression analysis is astatistical process for estimating therelationships among variables. It includesmany techniques for modeling and analyzingseveral variables, when the focus is on therelationship between a dependent variable andone or more independent variables.Eg:Predicting unemployment rate for nextyear.Estimatinginsurancepremium.those in other groups (clusters).eg: Finding customer segments in a companybased on transaction, web and customer calldata.Association analysisAssociation is a data mining function thatdiscovers the probability of the co-occurrenceof items in a collection. The relationshipsbetween co-occurring items are expressed asassociationrules.Eg: Find cross selling opportunities for aretailer based on transaction purchase history.Exploring and Expanding BusinessData mining is defined as a business processfor exploring large amounts of data to discovermeaningful patterns and rules [4]. Companiescan apply data mining in order to improvetheir business and gain advantages over thecompetitors. The most important businessareas that successfully apply data mining,presented in below figure.Anomaly detection:Anomaly detection (also outlier detection) isthe identification of items, events orobservations which do not conform to anexpected pattern or other items in a dataset.Eg:Fraud transaction detection in credit cards.Time seriesA time series is a series of data points indexed(or listed or graphed) in time order. Mostcommonly, a time series is a sequence taken atsuccessive equally spaced points in time. Thusit is a sequence of discrete-time data.Eg:Sales forecasting, production forecasting,virtually any growth phenomenon that needsto be extrapolated.Clusteringclustering is the task of grouping a set ofobjects in such a way that objects in the samegroup (called a cluster) are more similar (insome sense or another) to each other than toISSN: 2231-5381Digital marketing is a method of marketingfrom the viewpoint of customers. Digitalinformation is not only more easily integrated,sorted, and spread, but also enables providersand consumers to interact more quickly. In thepast, it usually took a long time for marketingto analyze and achieve effectiveness. Today,digital marketing enables marketing promotionto have a higher synergistic effect.With the rapid changes in the businessenvironment, technology progress, and digitaltransmission, the marketing of businessesshould be changed rapidly. Similarly, thehttp://www.ijettjournal.orgPage 181

International Journal of Engineering Trends and Technology (IJETT) – Special Issue – April 2017strategy of the market should be changed fromthe Red Sea Strategy to the Blue OceanStrategy. With the environment changing, notonly market space continues to expand, butalso the market environment becomes moretightened. From traditional store sales,telephone marketing, and face-to-facemarketing, to the development of Internet marketing, such as sale and purchase through theWebsite, keyword marketing, blog marketing,and so on, and with wireless networkdevelopment, the ubiquitous Internet will bebringing the world . It makes the developmentof digital marketing communications seemincreasingly important. The business is nolonger limited by traditional ways such as timeand space, but increases the opportunities ofcontact and interactions with customers.Apriori AlgorithmIn 1994, the Apriori algorithm was proposedby Agrawal and Srikant [3]. It is a classicalgorithm for learning association rules.Apriori is designed to operate on ions of items bought by customers, ordetails of commerce website frequentation).As is common in association rule mining,given a set of itemsets (for instance, sets ofretail transactions, each listing individualitems purchased), the algorithm attempts tofind subsets which are common to at least aminimum number of the item sets. Aprioriuses a “bottom up” approach, where frequentsubsets are extended one item at a time (a stepknown as candidate generation), and groups ofcandidates are tested against the data. Thealgorithm terminates when no furthersuccessful extensions are found.The purpose of the Apriori Algorithm is tofind associations between different sets ofdata. It is sometimes referred to as “MarketBasket Analysis”. There are a numberof items in each set, and is called a transaction.The output of Apriori is sets of rules that tellus how often items are contained in sets ofdata.III. Predicting the customer behaviourPredicting the customer behaviour is the mostimportant activity in enterprise business.Allthe previously described methods giveenterprises huge amounts of usefulinformation. the following section presentsISSN: 2231-5381some scenarios that have implemented theabove methods.Case study 1 : Application of Association Rulemining in Recommender systemsRecommender systems are hugely popularthese days in variety of fields. To name a fewMovies, music, books, research articles, searchqueries, social tags, etc. These systems helpthe enterprise by combining the ideas ofintelligent systems, machine learning,information retrieval to predict the customerbehaviour. There are two approaches inrecommender systems, one is collaborativefiltering and content-based filtering.Collaborative filtering methods collect andanalyze a large amount of information onusers’ behaviors, activities or preferences andpredict what users will like based on theirsimilarity to other users. One of theapproaches is to use the Apriori algorithm.In this case study Apriori algorithm is used toextract association rules from user profiles. Asan example PVT system is used. PVT systemis a recommender program that suggests TVchannels to users based on their viewinghabits. This system maintains both positivelyand negatively rated TV channels. Treatinguser profiles as transactions and the programratings therein as itemsets, the Apriorialgorithm can be used to derive a set of rulesand associated confidence levels betweenprograms. The confidence values are taken assimilarity scores and used to fill in a programsimilarity matrix. The procedure goes asfollows, the relationship between programs isidentified beyond a simple overlap. Like forexample a person who watches Reality showslike Rodies and Big Boss may not beinterested in shows like KBC and Indian Idol.But if a relation between Rodies and IndianIdol can be established then it can provide abasis for pattern matching. The relation can beidentified by finding the support andconfidence values. In this case study, theconfidence values are taken as the similarityscores and recommended to the user. Usingdirect program similarity we can derive rulesand further by chaining these rules together wecan get new results.Case Study 2: Classification model for Targetselection in direct marketingUsing historical purchase data, a predictiveresponse model with data mining techniqueshttp://www.ijettjournal.orgPage 182