A Role for People in the Machine Learning Algorithms of Predictive Analytics

by   |   October 3, 2012 4:27 pm   |   0 Comments

BOSTON- “I, for one, welcome our new computer overlords.”

That was all “Jeopardy!” champion Ken Jennings could say after he was bested by IBM’s Watson supercomputer in three straight days of quiz show competition in February 2011.

Using natural language processing techniques—with a system based on Hadoop—Watson could answer questions faster with more confidence than Jennings, who made a record 75 consecutive appearances on the show and made $2.5 million in winnings.

John Elder, the CEO and founder of Elder Research.

The computers haven’t taken over just yet—IBM is still tuning Watson for a full range of enterprise applications—but now more than ever is there an opportunity for humans to use the machines as detail-oriented assistants in business.

Related Stories

Podcast with Dean Abbott on the success of predictive analytics on the web.

Read more»

 

Utility project applies predictive analytics to slice of Pacific Northwest power grid.

Read more»

“The value of Watson is not coming up with the right answer, it’s coming up with all the possible answers and showing the evidence of why they could be right,” said Bob Jewell, the product director at IBM charged with making the supercomputer viable for business.

By showing a range of answers, scored by confidence level, and the evidence,  users can check Watson’s work and train the machine not to make the same mistakes. Users can add their own subject expertise and experience to make nuanced choices. This process of corrections and contributions serves to increase users’ confidence in the system, Jewell said.

Right now Watson is being used in health care by companies like WellPoint to help doctors treat patients, and by financial services companies to evaluate risk, but Jewell said IBM is making a push into other verticals.

Jewell and other speakers at Predictive Analytics World in Boston said technologies like machine learning, natural language processing and automatic hypothesis generation, where the computer suggests new research topic based on past results, work well enough today to do hours of research in milliseconds, freeing people up to ask better questions and  make contextual decisions.

John Elder, the founder and CEO of Elder Research, told attendees that data mining and predictive analytics have reached a point where researchers and analysts can confidently assign tedious, detail-oriented tasks to computers and focus on higher level decision making.

Collaborative Machines

Elder said today one of the most common use cases for data mining is fraud and anomaly detection, things that occur very infrequently but are a really big deal when they do.

“Analysts’ time is very precious, and you have a needle in a haystack problem, and you want to score all the hay for which ones are most likely to be needles,” Elder said. “That’s a really good use of data mining, because you can throw out 90 percent of the cases because there is zero chance of something being wrong, and then amongst the remaining 10 percent, you can prioritize them by risk. The analysts love that, because that gives them a to-do list.”

“If the computer has got some credibility is rating things for them, then they can use their judgment, but they don’t have that same terror that there is something out there big that they’re missing,” Elder said. “They feel that if it’s big enough and interesting enough then the computer would put it on their list.”

Peter Levine, the vice president of database marketing and yield management for Viking River Cruises, said his cruise company uses predictive analytics to better focus their direct marketing efforts. His company sends 60 million catalogues though the mail each year, he said, so making sure those offerings get in front of the people most likely to reserve a cabin (or a higher-priced suite) is crucially important.

“If I knew you were going to book at a higher dollar amount, I would spend more money to acquire you,” Levine said. Also valuable: predicting which customers will book trips directly and which will use a fee-based travel agent, he said. Such insights can improve the company’s profit margin.

Levine said his company has been using predictive analytics in-house for about four years; previous to that, he outsourced the work. Viking River Cruises has its own data, but also buys lists of prospective clients with special classifiers from third party companies.

Levine said he models how valuable the third party person data lists are to his company’s marketing efforts, measuring just how much value his vendors bring to the table.

“We get better and better as we get more data, and we push our vendors to get better,” he said. “One thing we’re also modeling is: what is the incremental value of company X’s data that they provide [to us]? It looks great, but what does it predict incrementally for what we already have?”

Email Staff Writer Ian B. Murphy at ian.murphy@wispubs.com. Follow him on Twitter .

 

Tags:

Post a Comment

Your email is never published nor shared. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>