Perform a literature review onEMERGINGTRENDSIN BUSINESSANALYTICS.Based on your review you need to submit an abstract in IEEE format.
Minimum Words:1500,
Minimum no of figures:3
Kohavi_ lo COMMUNICATIONS OF THE ACM August 2002/Vol. 45, No. 8 45 he field of business analytics has improved significantly over the past few years, giving business users better insights, particularly from operational data stored in transactional systems. An example is e-com- merce data analysis, which has recently come to be viewed as a killer app for the field of data mining [5, 6]. The data sets cre- ated by integrating clickstream records generated by Web site activity with demographic and other behav- ioral data dwarf, in size and complexity, the largest data warehouses of just a few years ago [4]. The result is massive databases requiring a mix of automated analysis techniques and human effort to give business users strategic insight about the activity on their sites, as well as about the characteristics of the sites’ visitors and customers. With many millions of clickstream records generated every day, aggregated to customer- focused records with hundreds of attributes, there is a clear need for automated techniques for finding pat- terns in the data. Here, we discuss the technology and enterprise-adoption trends associated with business analytics. The key consumer is the business user, whose job, possibly in merchandising, marketing, or sales, is not directly related to analytics per se, but who typically uses analytical tools to improve the results of some business process along one or more dimensions (such as profit and time to market). Fortunately, data min- ing,1 analytic applications, and business intelligence systems are now better integrated with transactional systems than they once were, creating a closed loop between operations and analysis that allows data to be analyzed and the results reflected quickly in busi- ness actions. Mined information today is deployed to a broader business audience taking advantage of business analytics in its everyday activities. Analytics are now routinely used in sales, marketing, supply chain optimization, and fraud detection [2, 3]. Business Users Even with these advances, business users, while expert in their particular areas, are still unlikely to be expert in data analysis and statistics. To make deci- sions based on the data collected by and about their organizations, they must either rely on data analysts to extract information from the data or employ ana- lytic applications that blend data analysis technolo- gies with task-specific knowledge. In the former, business users impart domain knowledge to the ana- lyst, then wait for the analyst to organize and analyze it and communicate back the results. These results typically raise further questions, hence several itera- tions are necessary before business users can actually act on the analysis. In the latter, analytic applications incorporate not only a variety of data mining tech- niques but provide recommendations to business users as to how to best analyze the data and present the extracted information. Business users are expected to use it to improve performance along multiple metrics. Unfortunately, the gap between relevant analytics and users’ strategic business needs is significant. The gap is characterized by several challenges: The goal is business effectiveness through ‘verticalization,’ usability, and integration with operational systems. EMERGING TRENDS BY RON KOHAVI, NEAL J. ROTHLEDER, AND EVANGELOS SIMOUDIS IN BUSINESS ANALYTICS 1The terms data mining and analytics are used interchangeably here to denote the general process of exploration and analysis of data to discover new and meaningful patterns in data. This definition is similar to those in [2, 3] where it’s referred to as knowledge discovery. Cycle time. The time needed for the overall cycle of collecting, analyzing, and acting on enterprise data must be reduced. While business constraints may impose limits on reducing the overall cycle time, business users want to be empowered and rely less on other people to help with these tasks. Analytic time and expertise. Within the overall cycle, the time and analytic expertise necessary to analyze data must be reduced. Business goals and metrics. Unrealistic expectations about data mining “magic” often lead to misguided efforts lacking clear goals and metrics. Goals for data collection and transformations. Once metrics are identified, organizations must collect and transform the appropriate data. Data analysis is often an afterthought, limiting the possible value of any analysis. Distributing analysis results. Most analysis tools are designed for quantitative analysts, not the broader base of business users who need the output trans- lated into language and visualizations appropriate for business needs. Integrating data from multiple sources. The extract- transform-load (ETL) process is typically complex, and its cost and difficulty are often underestimated. The Driving Force The emerging trends and innovations in business ana- lytics embody approaches to these business challenges. Indeed, it is a very healthy sign for the field that regardless of the solution-process, technology, system integration, or user interface, business problems remain the driving force. “Verticalization.” In order to reduce discovery cycle time, facilitate the definition and achievement of busi- ness goals, and deploy analysis results to a wider audi- ence, developers of analytical solutions started verticalizing their software, or customizing applica- tions within specific industries. The first step toward verticalization was to incorporate task-specific knowl- edge; examples include: knowledge about how to ana- lyze customer data to determine the effectiveness of a marketing campaign; knowledge of how to analyze clickstream data generated by a Web site to reduce shopping cart abandonment and improve ad effective- ness; knowledge about how an investment bank con- solidates its general ledger and produces various types of forecasts; and how an insurance company analyzes data in order to provide an optimally priced policy to an existing customer. In the process of incorporating industry-specific knowledge, companies are also able to optimize the performance of their applications for specific indus- tries. For example, a company that developed an ana- lytic application for budgeting and forecasting targeted at the financial services industry determined that its online analytical processing, or OLAP, engine’s execu- tion speed could be optimized by limiting to nine the number of dimensions it had to handle, a number deemed sufficient for the particular application in that industry. The use of industry-specific knowledge is not lim- ited to the data mining components of analytic appli- cations but also affects how the extracted information is accessed and presented. For example, organizations in the financial services, retail, manufacturing, utilities, and telecommunications industries increasingly want their field personnel to have access to business analytic information through wireless devices. Analytic appli- cation vendors are now developing technologies to automatically detect wireless devices and their form 46 August 2002/Vol. 45, No. 8 COMMUNICATIONS OF THE ACM A visualization of a Naive Bayes model for predicting who in the U.S. earns more than $50,000 in yearly salary. The higher the bar, the greater the amount of evidence a person with this attribute value earns a high salary. factors, automatically tailoring analysis results to fit the capabilities of a particular device. For example, if the information is to be displayed on a phone supporting the Wireless Access Protocol (implying small screen size), it may be necessary to automatically summarize text, abbreviate words, and limit the use of graphics by automatically selecting only the most relevant figures. Comprehensible models and transformations. In light of the need to let business users analyze data and quickly gain insight, and aiming for the goal of reduc- ing reliance on data mining experts, comprehensible models are more popular than opaque models. For example, in the KDD-Cup 2000 [5], a data mining competition in which insight was important, the use of decision trees, generally accepted as relatively easy to understand, outnumbered other methods more than two to one. Business users do not want to deal with advanced statistical concepts; they want straightforward visual- izations and task-relevant outputs. The figure outlines a Naive Bayes model for predicting who in the U.S. earns more than $50,000 in yearly salary. Instead of the underlying log conditional probabilities the model actually manipulates, the visualization uses bar height to represent evidence for each value of a contributing factor listed on the left and color saturation to signify confidence of that evidence [1]. For example, evidence for higher salaries increases with age, until the last age bracket, when it drops off; evidence for higher salaries increases with years of education, number of hours worked, and certain marital status and occupations. Note also the visualization shows only a few attributes determined by the mining algorithm to be the most important ones, highlighting to business users the most critical attributes from a larger set. Other exam- ples of visualizing data and data mining models are in [7, 9]. Part of the larger system. The needs of data analysis are being designed into systems, instead of being an afterthought, typically addressing the following areas: Data collection. You cannot analyze what you do not collect, so collecting rich data is critical. For exam- ple, e-commerce systems can collect attributes rang- ing from the user’s local time, screen resolution (useful for determining the quality of images to send), and network bandwidth. Generation (and storage) of unique identifiers. In order to help merge information from several records and remove duplicate records, systems must generate unique keys to join data and store them. For exam- ple, all clickstream records in the same session should store the session IDs so they can be joined later to session records stored in other tables. Integration with multiple data sources. Analysis is more effective when data is available from multiple sources. For example, in customer analytics, data should be merged from multiple touchpoints, including the Web, call centers, physical stores, wireless access, and ad campaigns (both direct and online). Behavioral data can be more powerful when overlaid with demographic and socioeco- nomic data from other sources. Hardware sizing. Analysis requires hardware capable of dealing with large amounts of data. Some organi- zations have traditionally underestimated the need for sophisticated IT infrastructure and the hardware needed to make timely analysis feasible. In new areas. During the past few years, recognition of the strategic value of business analytics has led to sig- nificant developments in business applications that analyze customer data. They’ve been used to reduce customer attrition, improve customer profitability, increase the value of e-commerce purchases, and increase the response of direct mail and email market- ing campaigns. This success has paved the way for new applica- tions; three are particularly promising: supply chain visibility, price optimization, and work force analysis. Organizations have automated portions of their supply chains, enabling collection of significant data about inventory, supplier performance, and logistics of mate- rials and finished goods. Newer applications analyze this data to provide insights about the performance of suppliers and partners, material expenditures, accuracy of sales forecasts for controlling materials inventory, accuracy of production plans, and accuracy of plans for order delivery. The wide adoption of customer relationship man- agement, or