Data Mining Case Studies

Data mining application at present in the domestic basic conclusion is "large enterprise success cases, small and medium-sized enterprises need less." But for the market, if it is not really "no one to buy" so "no one to sell", it must be the opportunity for innovation. Personal judgment is that a database as long as more than hundreds of thousands of records, there is the value of data mining.

Collect the following cases, hope to have some inspiration and learning value.


1. Which products are better sold together?

This is Wal-Mart's classic case: beer and diapers are generally a completely different product of the customer base. But the results of Wal-Mart's data mining within a year suggest that the store's beer, which sells well in residential areas, sells well. The reason is actually very simple, the general wife lets the gentleman downstairs buys the diaper, the gentleman generally will treat oneself two to listen to the beer. So beer and diapers are the most opportunities to buy together. This is a modern mall intelligent information Analysis system found the secret. This story is recognized as the birth of data mining in the business world.

In addition, we all know that there is usually a toothbrush next to Wal-Mart toothpaste, which is placed on the price so that toothpaste and toothbrush can be sold well.

2. Inventory Forecast

In the past, retailers relied on supply chain software, internal analysis software, and even intuition to predict inventory requirements. As competitive pressures increase day by day, many retailers, from major finance executives to inventory managers, are working to find more accurate ways to predict the inventory their chain stores should hold. Predictive analytics is a solution. It can accurately predict which store locations should keep what products.

Using Analysis Services in Microsoft (R) SQL Server (TM) 2005 and SQL Server Data Warehouse, data mining technology is used to provide accurate and timely information for product storage decisions. The data mining model that SQL Server 2005 Analysis Services obtains can predict whether a book will be out of stock in the next week, with an accuracy of 98.52%. On average, the accuracy of predicting whether the book will be out of stock in the next two weeks is 86.45%.

3. Stock Presets

Predicting the movement of a stock is almost impossible, but through correlation analysis, you can find out the trend of a stock and the potential rules of another stock trend, such as data mining once got this conclusion: "If Microsoft's shares fell 4%, then IBM's shares will fall 5% in two weeks."

4. How do NBA coaches work to improve their chances of winning?

The NBA coach of America's famous national basketball team uses data mining tools provided by IBM to decide to replace players. Imagine you are an NBA coach, what do you do to lead your team to victory? Of course, the most easily thought of is the full-court press, cross-pull and quick steals and other specific tactics and techniques. But today, NBA coaches have their new weapon: data mining. About 20 NBA teams used IBM's developed data Mining Application Advanced Scout system to optimize their tactical mix. For example, Scout has found a chance to win in the Miami Heat team because he studied the different arrangements of the Magic team.

System analysis showed that the two defenders of the Magic team Anforni Hadwich (Anfernee Hardaway) and brother Branham (Brian Shaw) were rated as 17 points in the first two games, which meant that the team lost more than 17 points on the pitch. However, when Hadwich with the backup defender Daryl Armstrong (Darrell Armstrong), the Magic team scored positive 14 points.

In the next game, the Magic team increased the playing time of Armstrong. It worked: Armstrong scored 21 points, Hadwich scored 42 points and the Magic team won 88:79. In the fourth game, the Magic team got Armstrong into the first lineup and once again defeated the Heat team. In the fifth game, the data-mining-backed lineup failed to stall the heat, but advanced Scout helped the Magic team win 5 games, until the end.

Advanced Scout is a data analysis tool that coaches can use laptops at home or on the road to mine data stored on servers in the NBA center. Each event is classified by statistics, by score, assists, mistakes, and so on. Time markers make it easy for coaches to understand the meaning of statistical discoveries by searching for videos of NBA games. For example, the coach discovers from advanced Scout that the player has a foul record when opposing a star, and that he can break down the interaction between the opposing star and the player's head, and then design a reasonable defensive strategy.

The developer of Advanced Scout, Departout Buhandri, who developed the application when he was a researcher at IBM's Thomasj.watson Research Center, demonstrated how a novice technology should use data mining. "Coaches can have no statistical training at all, but they can use data mining to develop strategies," Buhandri said. Meanwhile, another official sports league, the National Hockey League, is developing its own data mining application Nhl-ice, which has established a technology-based joint venture with IBM, which launched an electronic real-time scoring and statistics system last November. In principle, it is a data mining application similar to Advanced Scout that allows coaches, broadcasters, journalists and fans to dig up the NHL stats. When they visit the NHL web site, fans can use the system to cycle through the league matches, while broadcasters and journalists can dig up statistics and find lace news to spice up their live commentary.

5. Out of a new product, which old customers are most likely to buy?

The Bank of Montreal is Canada's oldest bank and Canada's third largest bank. In the mid 1990s, the industry's increased competition led to a need for the bank to target 18 million customers through cross-selling. "This reflects a new focus of the bank – the customer (not the commodity)," said Mrazek, senior manager of bank intelligence business. Banks should recognize what products customers need and how to market them, rather than waiting for people to queue up to buy them. The bank then needs to develop the goods and conduct marketing campaigns to meet these needs.

Before applying data mining, the bank's sales representative must sell the product to the customer in a specific area from 6 o'clock to 9 o ' night. But, as everyone on the receiving end knows, most people are not interested in peddling after the end of their work. Therefore, the feedback rate for telemarketing at dinner time is very low.

A few years ago, the bank began using IBM DB2 Intelligent Miner scoring to evaluate records based on bank account balances, customer-owned banking products, and location and credit risk standards. These reviews can be used to determine the likelihood of a customer buying a specific product. The system can be viewed through a browser window, so that managers do not have to analyze the underlying data, making it ideal for non-statisticians.

"We have a deeper understanding of our clients ' financial behaviour habits and their impact on bank profitability. Now, when conducting more targeted marketing campaigns, banks can differentiate between different customer groups to improve the quality of their products and services, while at the same time setting the right price, designing various incentive schemes, and even determining interest charges. "

Bank of Montreal Data mining tools provide managers with a wealth of information to help them make decisions about everything from marketing to product design.

6. What is the most likely purchase behavior on public pages of e-commerce sites?

Santiago's Proflowers.com uses Hitbox, the WebSideStory Data Mining ASP service, to enable business planners to respond quickly to sales on peak days. Because the flowers are very easy to wither, proflowers have to evenly cut inventory, otherwise it may lead to a commodity too quickly sold out or inventory of flowers wither.

As the daily trading volume is high, managers need to analyze the retail situation, such as the conversion rate, that is, how much page views will lead to sales. For example, if only 5 of the 100 people see Roses and the Bonsai conversion rate is 100:20, then there is no problem with the page design, which is the price of the roses. The company can quickly adjust the site, such as showing roses on every page or lowering the price of roses. For goods that may be sold out too quickly, companies often have to weaken the product or cancel the offer in a webpage to try to slow the sale of the product.

The advantage of using Hitbox is that it shows sales data and conversion rates with easy-to-read displays. Chris d ' Eon, vice president of ProFlowers Marketing, said: "It's a waste of time to analyze your data." We need a way to navigate the data that will allow us to take immediate action. "

7. What is the most likely purchase for the current user who is logged into the site?

Ebags in Denver aims to sell suitcases, handbags, wallets and other travel services for frequent flyers. The company uses Kana Software's e-marketing Suite to integrate its website's Oracle database, J.D. Edwards financial system, customer service e-mail and call center to get information on customer buying behavior habits. Data analysis can help companies determine which page is causing the customer's high purchasing rate and understand what is driving the sale.

"We try to show different things to see what's best for promotion," said Mike Frazini, vice president of Ebags Technology. Our ultimate goal is to be completely personalized. "Unlike the design page to encourage most consumer purchases, a personalized solution will continue to create pages to fit every specific visitor. Therefore, if a visitor's browsing history shows that it is interested in the handbag, the site creates a customized page that highlights these items. Frazini points out that the analysis methods used to implement data mining today can also be used to deploy automated site customization rules.

Finding a personalized Web page based on less data and business rules is one of the ways that a customer-specific site can reduce resource costs. Carrier, an air-conditioning manufacturer in Connecticut State Farmington, USA, claims that the average revenue generated by each visitor to its upgraded business-to-consumer site has increased from $1.47 to $37.42 in one months, just by using zip code data.

When customers sign in to the site, they are instructed to provide a postal code. These zip code information will be sent to the Webminer server, which is a data mining ASP. Webminer's data mining software then assumes the customer and displays the product based on these assumptions. For example, if the customer is from a wealthy suburban area, the website will show an air conditioner with a remote control, and if the customer's ZIP code shows a large number of apartment buildings nearby, pop-up ads will show window-type air conditioners.

By using this relatively simple approach, the company can generate Web pages in seconds. Carrier, global e-commerce manager Paul Berman said: "Contrary to the usual idea, customer-oriented ecommerce does not need to ask customers 8 or 9 of information when creating targeted services. We only need 1 messages, and the actual results are really good. "

Like Carrier, the "friend of the musician" (musician's friend) is also reducing the business rules used to determine customer-defined content. It is a directory of Guitar Center Limited and a WEB branch office.

Data Mining Case Studies