Predictive Analytics (PA)
By Thomas H. Davenport
Introduction
People Consume
> Hollywood studios predict the success of a screenplay if produced.
> Wall Street predicts stock prices by observing how demand drives them up and down.
> Companies predict which customer will buy their products in order to target their marketing. The prediction dictate the allocations of marketing budgets. Some companies predict how to best influence you to buy more.
> Predicting mouse clicks pays off massively
People Love, Work, Procreate, and Divorce
> The leading career-focused social network, LinkedIn, predicts your job skills.
> Online dating predict which hottie on your screen would be the best bet at your side.
> Target predicts customer pregnancy in order to market relevant products accordingly.
> Clinical researchers predict infidelity and divorce. Credit card companies do the same.
People Think and Decide
> Obama was re-elected in 2012 with the help of voter prediction. The Obama for America Campaign predicted which voters would be positively persuaded by campaign contact (a call, door knock, flier, or TV ad) and which would actually be inadvertently influenced to vote adversely by contact. Employed to drive campaign decision for millions of swing state voters, this method was shown to successfully convince more voters to choose Obama than traditional campaign targeting.
> "What did you mean by that>" Systems have learned to ascertain the intent behind the written word. Citibank and PayPal detect the customer sentiment about their products, and one researcher's machine can tell which Amazon.com book reviews are sarcastic.
> Student essay grade prediction has been developed for possible use to automatically grade. The system grades as accurately as human graders.
> There's a machine that can participate in the same capacity as humans in US's most popular broadcast celebration of human knowledge and cultural literacy. This machine learned to work proficiently enough with English to predict the answer to free-form inquiries across an open range of topics and defeat the two all-time human champs.
> Computers can literally read your mind. Researchers trained systems to decode a scan of your brain and determine which type of object you're thinking about -- such as certain tools, buildings, and food -- with over 80 percent accuracy for some human subjects. In 2011, IBM predicted that mind-reading technology would be mainstream within five years.
People Quit
> Hewlett-Packard earmarks each and every one of its more than 330,000 worldwide employees according to "Flight Risk", the expected chance he/she will quit their job so that managers may intervene in advance where possible, and plan accordingly otherwise.
> Ever experience frustration with your cell phone service? Your service provider endeavors to know. All major wireless carries predict how likely it is you will cancel and defect to a competitor -- possibly before you have conceived a plan to do so -- based on factors such as dropped calls, your phone usage, billing information, and whether you contacts have already defected.
> FedEx predicts 65 to 90 percent accuracy which customers are at risk of defecting to acompetitor.
> Wikipedia predicts which of its editors who work for free as a labor of love to keep this priceless online asset alive, are going to discontinue their valuable service.
> Researchers at Harvard Medical School predict that if your friends stop smoking, you're more likely to do so yourself as well. Quitting smoking is contagious.
People Mess Up
> Insurance companies predict who is going to carsh a car or take a bad ski jump.
> For is learning from data so its cars can detect when the driver is not alert due to distraction, fatigue or intoxication and take action such as sounding an alarm.
> Researchers have identified aviation incidents that are five times more likely than average to be fatal, using data from the National Transportation Safety Board.
> All large banks and credit card companies predict which debtors are most likely to turn delinquent, failing to pay back their loans or credit card balances. Collection agencies prioritize their efforts with predictions of which tactic has the best chance to recoup the most from each defaulting debtor.
People Get Sick and Die
> Heritage Provider Network made competition for scientists in predicting individual hospital admission. By following these predictions, proactive preventive measures can take a healthier bite out of the tens of billions of dollars spent annually on unnecessary hospitalization.
> Life insurance companies go beyond conventional actuarial tables and employ predictive technology to establish mortality risk. Int's not called death insurance, but they calculate when you are going to die.
> Doctors, unintentionally, sacrifice some patients in order to save others, and this is done completely without controversy, by predicting something besides diagnosis or outcome: healthcare impact.
People Lie, Cheat, Steal and Kill
> Most banks employ predictive technology to counter the assault of fraudulent checks, credit card charges, and other transactions.
> Predictive computers help decide who belongs in prison.
> Murder is widely considered impossible to predict accurately, but within at-risk population predictive methods can be effective.
> A fraud expert at a large UK bank extended his work to discover a small pool of terror suspects based on their banking activities.
> Police patrol the areas predicted to spring up as crime hot spots.
> Researchers trained a system to detect lies with 82% accuracy by observing eye movements alone.
> A university research team employed cheating detection software to patrol hundreds of computer programming homework submission for plagiarism.
> The IRS predicts if you are cheating on your taxes.
The Limits and Potential of Prediction
> The Data of Crime and The Crime of Data
Oregon launched a crime prediction tool to be consulted by judges when sentencing convicted felons. If you know convict's state ID and the crime for which he/she is being sentenced, you can enter the website of Oregon Criminal Justice Commission and see the predictive model's output: the probability the offender will be convicted again for a felony within three years of being released. Over half of these offenders will commit a felony again. Studies have shown that arbitrary extraneous factors greatly affect judicial decisions. Hungry judges rule negatively. If your parole board judges are hungry, you're more likely to stay in prison. With the predicting machine, convict's future now rest in nonhuman hands. Given a new power, the computer can commit more than just prediction errors -- it can commit injustice, previously a form of misjudgment that only people were in a position to make. After all, the price is not as high when an e-mail message is wrongly incarcerated in the spam folder or a fraud auditor's time is wasted on a transaction that turns out to be legitimate.
> Machine Risk Without Measure
In "Minority Report", the cop tackles and handcuffs individuals who have committed no crime (yet). Rather than punishment fitting the crime, the punishment fits the precrime. A false positive, aka false alarm, is when a model incorrectly predicts yes, when the correct answer is no. The risk of injustice is nothing new, since human parole boards and judges face the same problem as they regularly make predictions about criminals' future behavior. What is new here, despite a general movement toward upgrading decision making with data, is entrusting a machine to contribute to these life-changing decisions for which there can be no accountability. Security is often at odds with civil liberties. The act of balancing between the two gets even trickier with predictive technology at play.
> The Cyclicity of Prejudice
Yet another quandary lurks. Although science promises to improve the effectiveness and efficiency of law enforcement, when you formalize and quantify decision making, you inadvertently instill existing prejudices against minorities. Why? Because prejudice is cyclic, a self-fulfilling prophesy, and this cycling could be intensified by PA's deployment.
> Good Prediction, Bad Prediction
Pregnancy prediction faces the opposite dilemma of that faced by crime prediction. Crime prediction causes damage when it predicts wrong, but predicting sensitive facts like pregnancy can cause damage when it's right. Knowledge of a pregnancy is extremely potent, and leaking it to the wrong ears can be life-changing indeed. Imagine the pregnant woman's job is shaky. Sometimes it's better not to know, like the idea of predicting employee death. Predicting death is so sensitive that it's done secretly, keeping it on the down low even when done for benevolent purposes.
Put your money where your mouth is
The one thing all human beings do when they're confronted with uncertainty is pull back, withdraw, disengage -- and that means economic activity just goes straight down. - Alan Greenspan
Money is singular measure of how people are faring, so can't we expect our emotional and financial well-being to be closely tied? One could say that everything comes down to feelings, even money. More cynically, you might ask whether its' actually the other way around. The stock market served as an ideal stomping ground within which to validate the Anxiety Index. To empirically resolve the chicken-and-egg dilemma of whether emotion "hatches" action or the other way around, the economy could serve as an established standard from which to observe the optimistic and pessimistic fluctuations of society as a whole. Beyond scientific validation, a tantalizing prospect lingered: stock market prediction. If collective emotion proved to be reflected by subsequent stock movements, the blog mood readings could serve to predict them.
THE DATA EFFECT = Data is always predictive.
Data always speak. It always has a story to tell. Pull some data together, although you can never be certain what you'll find, you can be sure you'll discover valuable connections by decoding the language it speaks and listening. That's the Data Effect in a nutshell.
Bizarre and Surprising Insights -- Consumer Behavior
1. Guys literally drool over sports cars. Male college student subjects produce more saliva when presented with images of sports cars or money. Consumer impulses are physiological cousins of hunger.
2. If you buy diapers, you are more likely to also buy beer. Daddy needs a beer.
3. Dolls and candy bars. Customers who buy a barbie doll also buy candy bars. Kids come along for errands.
4. Staplers reveal hires. The purchase of a staler often accompanies the purchase of paper, waste baskets, scissors, paper clips, folders and so on.
5. Mac users book more expensive hotels.
6. Your inclination to buy varies by time of day. For retail websites the peak is 8 pm, for dating late at night, for finance around 1 pm, for travel just after 10 am.
7. Your e-mail address reveals your level of commitment.
8. Banner ads affect you more than you think. Advertising exerts a subconscious effect.
9. Companies win by not prompting customers to think.
10. Your web browsing reveals yur intentions.
11. Friends stick to the same cell phone company (a social effect).
12. Low credit rating, more car accidents.
13. Your shopping habits foretell your reliability as a debtor.
14. Small businesses' credit risk depends on the owner's behavior as a consumer.
15. Within a certain genetics cluster, having more genes shared by heterosexual couple means more infidelity by the female. We're programmed to avoid inbreeding, since there are benefits to genetic diversity.
16. Retirement is bad for your health, unhealthy habits such as smoking and drinking follow retirement.
17. Google search (Google Flu Trends) tends to predict disease outbreak. People with symptoms or in the vicinity of others with symptoms seek further information.
18. Smokers suffer less from repetitive motion disorder, because smokers take more breaks.
19. Positive health habits are contagious. People are strongly influenced by their social environment.
20. Happiness is contagious (a social effect). Each additional Facebook friend who is happy increases your chances of being happy by roughly 9%. Waves of happiness spread throughout the network.
21. Music expedites poststroke recovery and improves mood. Music listening activates a widespread bilateral network of brain regions related to attention, semantic processing, memory, motor functions and emotional processing.
22. Yoga improves your mood. Yoga is designed for, and practiced with the intent for, the attainment of tranquility.
23. Suicide bombers do not buy life insurance.
24. Unlike lighting, crime strikes twice.
25 Crime rises with public sporting events.
26. Crime rises after elections.
27. Music taste predicts political affiliations.
28. A job promotion can lead to quitting.
29. Vegetarians miss fewer flights.
30. Solo rockers die younger than those in bands.
Serendipity and Innovation
If necessity is the mother of invention, serendipity is its daddy. It was only by happy accident that Alexander Fleming happened upon the potent effects of penicillin, by noticing that an old bacterial culture he was about to clean up happened to be contaminated with some mold -- which was successfully killing it. By its very design, PA fosters serendipity. Predictive modeling conducts a broad, exploratory analysis, testing many predictors, and in so doing uncovers surprising findings, such as vegetarians being less likely to miss flights and so on.
The Ensemble Effect
(Netflix, CrowdSourcing)
Movie recommendations:
1. What's predicted: what rating a customer would give to a movie
2. What's done about it: customers are recommended movies that they are predicted to rate highly
The collective intelligence of a crowd emerges on many occasions:
1. Prediction markets, wherein a group of people together estimate the prospects for a horse race, political event, or economic occurrence by way of placing bets.
2. The audience of TV quiz show Who Wants to Be A Millionaire, whom contestants may poll to weigh in on questions.
3. Google's PageRank method, by which a web page's value and importance are informed by how many links people have created to point to the page.
4. The predictive capacity of the mass mood expressed by bloggers at large to foresee stock market behavior.
Human minds aren't the only things that can be effectively merged together. It turns out the aggregate effect emerging from a group extends also to nonhuman crowds -- of predictive models.
Like a crowd of people, an ensemble of predictive models benefits from the same "collective intelligence" effect. A Bag of Models = to predict, each model make its prediction and tally up the results. The Generalization Paradox: More is Less. When joined in an ensemble, predictive models compensate for one another's limitations, so the ensemble as w hole is more likley to predict correctly than its component models are.
A New Thing To Predict
Targeted Marketing with Response Modeling
1. What's predicted: which customers will purchase if contacted
2. What's done about it: contact those customers who are more likely to do so.
Introduction
People Consume
> Hollywood studios predict the success of a screenplay if produced.
> Wall Street predicts stock prices by observing how demand drives them up and down.
> Companies predict which customer will buy their products in order to target their marketing. The prediction dictate the allocations of marketing budgets. Some companies predict how to best influence you to buy more.
> Predicting mouse clicks pays off massively
People Love, Work, Procreate, and Divorce
> The leading career-focused social network, LinkedIn, predicts your job skills.
> Online dating predict which hottie on your screen would be the best bet at your side.
> Target predicts customer pregnancy in order to market relevant products accordingly.
> Clinical researchers predict infidelity and divorce. Credit card companies do the same.
People Think and Decide
> Obama was re-elected in 2012 with the help of voter prediction. The Obama for America Campaign predicted which voters would be positively persuaded by campaign contact (a call, door knock, flier, or TV ad) and which would actually be inadvertently influenced to vote adversely by contact. Employed to drive campaign decision for millions of swing state voters, this method was shown to successfully convince more voters to choose Obama than traditional campaign targeting.
> "What did you mean by that>" Systems have learned to ascertain the intent behind the written word. Citibank and PayPal detect the customer sentiment about their products, and one researcher's machine can tell which Amazon.com book reviews are sarcastic.
> Student essay grade prediction has been developed for possible use to automatically grade. The system grades as accurately as human graders.
> There's a machine that can participate in the same capacity as humans in US's most popular broadcast celebration of human knowledge and cultural literacy. This machine learned to work proficiently enough with English to predict the answer to free-form inquiries across an open range of topics and defeat the two all-time human champs.
> Computers can literally read your mind. Researchers trained systems to decode a scan of your brain and determine which type of object you're thinking about -- such as certain tools, buildings, and food -- with over 80 percent accuracy for some human subjects. In 2011, IBM predicted that mind-reading technology would be mainstream within five years.
People Quit
> Hewlett-Packard earmarks each and every one of its more than 330,000 worldwide employees according to "Flight Risk", the expected chance he/she will quit their job so that managers may intervene in advance where possible, and plan accordingly otherwise.
> Ever experience frustration with your cell phone service? Your service provider endeavors to know. All major wireless carries predict how likely it is you will cancel and defect to a competitor -- possibly before you have conceived a plan to do so -- based on factors such as dropped calls, your phone usage, billing information, and whether you contacts have already defected.
> FedEx predicts 65 to 90 percent accuracy which customers are at risk of defecting to acompetitor.
> Wikipedia predicts which of its editors who work for free as a labor of love to keep this priceless online asset alive, are going to discontinue their valuable service.
> Researchers at Harvard Medical School predict that if your friends stop smoking, you're more likely to do so yourself as well. Quitting smoking is contagious.
People Mess Up
> Insurance companies predict who is going to carsh a car or take a bad ski jump.
> For is learning from data so its cars can detect when the driver is not alert due to distraction, fatigue or intoxication and take action such as sounding an alarm.
> Researchers have identified aviation incidents that are five times more likely than average to be fatal, using data from the National Transportation Safety Board.
> All large banks and credit card companies predict which debtors are most likely to turn delinquent, failing to pay back their loans or credit card balances. Collection agencies prioritize their efforts with predictions of which tactic has the best chance to recoup the most from each defaulting debtor.
People Get Sick and Die
> Heritage Provider Network made competition for scientists in predicting individual hospital admission. By following these predictions, proactive preventive measures can take a healthier bite out of the tens of billions of dollars spent annually on unnecessary hospitalization.
> Life insurance companies go beyond conventional actuarial tables and employ predictive technology to establish mortality risk. Int's not called death insurance, but they calculate when you are going to die.
> Doctors, unintentionally, sacrifice some patients in order to save others, and this is done completely without controversy, by predicting something besides diagnosis or outcome: healthcare impact.
People Lie, Cheat, Steal and Kill
> Most banks employ predictive technology to counter the assault of fraudulent checks, credit card charges, and other transactions.
> Predictive computers help decide who belongs in prison.
> Murder is widely considered impossible to predict accurately, but within at-risk population predictive methods can be effective.
> A fraud expert at a large UK bank extended his work to discover a small pool of terror suspects based on their banking activities.
> Police patrol the areas predicted to spring up as crime hot spots.
> Researchers trained a system to detect lies with 82% accuracy by observing eye movements alone.
> A university research team employed cheating detection software to patrol hundreds of computer programming homework submission for plagiarism.
> The IRS predicts if you are cheating on your taxes.
The Limits and Potential of Prediction
> The Data of Crime and The Crime of Data
Oregon launched a crime prediction tool to be consulted by judges when sentencing convicted felons. If you know convict's state ID and the crime for which he/she is being sentenced, you can enter the website of Oregon Criminal Justice Commission and see the predictive model's output: the probability the offender will be convicted again for a felony within three years of being released. Over half of these offenders will commit a felony again. Studies have shown that arbitrary extraneous factors greatly affect judicial decisions. Hungry judges rule negatively. If your parole board judges are hungry, you're more likely to stay in prison. With the predicting machine, convict's future now rest in nonhuman hands. Given a new power, the computer can commit more than just prediction errors -- it can commit injustice, previously a form of misjudgment that only people were in a position to make. After all, the price is not as high when an e-mail message is wrongly incarcerated in the spam folder or a fraud auditor's time is wasted on a transaction that turns out to be legitimate.
> Machine Risk Without Measure
In "Minority Report", the cop tackles and handcuffs individuals who have committed no crime (yet). Rather than punishment fitting the crime, the punishment fits the precrime. A false positive, aka false alarm, is when a model incorrectly predicts yes, when the correct answer is no. The risk of injustice is nothing new, since human parole boards and judges face the same problem as they regularly make predictions about criminals' future behavior. What is new here, despite a general movement toward upgrading decision making with data, is entrusting a machine to contribute to these life-changing decisions for which there can be no accountability. Security is often at odds with civil liberties. The act of balancing between the two gets even trickier with predictive technology at play.
> The Cyclicity of Prejudice
Yet another quandary lurks. Although science promises to improve the effectiveness and efficiency of law enforcement, when you formalize and quantify decision making, you inadvertently instill existing prejudices against minorities. Why? Because prejudice is cyclic, a self-fulfilling prophesy, and this cycling could be intensified by PA's deployment.
> Good Prediction, Bad Prediction
Pregnancy prediction faces the opposite dilemma of that faced by crime prediction. Crime prediction causes damage when it predicts wrong, but predicting sensitive facts like pregnancy can cause damage when it's right. Knowledge of a pregnancy is extremely potent, and leaking it to the wrong ears can be life-changing indeed. Imagine the pregnant woman's job is shaky. Sometimes it's better not to know, like the idea of predicting employee death. Predicting death is so sensitive that it's done secretly, keeping it on the down low even when done for benevolent purposes.
Put your money where your mouth is
The one thing all human beings do when they're confronted with uncertainty is pull back, withdraw, disengage -- and that means economic activity just goes straight down. - Alan Greenspan
Money is singular measure of how people are faring, so can't we expect our emotional and financial well-being to be closely tied? One could say that everything comes down to feelings, even money. More cynically, you might ask whether its' actually the other way around. The stock market served as an ideal stomping ground within which to validate the Anxiety Index. To empirically resolve the chicken-and-egg dilemma of whether emotion "hatches" action or the other way around, the economy could serve as an established standard from which to observe the optimistic and pessimistic fluctuations of society as a whole. Beyond scientific validation, a tantalizing prospect lingered: stock market prediction. If collective emotion proved to be reflected by subsequent stock movements, the blog mood readings could serve to predict them.
THE DATA EFFECT = Data is always predictive.
Data always speak. It always has a story to tell. Pull some data together, although you can never be certain what you'll find, you can be sure you'll discover valuable connections by decoding the language it speaks and listening. That's the Data Effect in a nutshell.
Bizarre and Surprising Insights -- Consumer Behavior
1. Guys literally drool over sports cars. Male college student subjects produce more saliva when presented with images of sports cars or money. Consumer impulses are physiological cousins of hunger.
2. If you buy diapers, you are more likely to also buy beer. Daddy needs a beer.
3. Dolls and candy bars. Customers who buy a barbie doll also buy candy bars. Kids come along for errands.
4. Staplers reveal hires. The purchase of a staler often accompanies the purchase of paper, waste baskets, scissors, paper clips, folders and so on.
5. Mac users book more expensive hotels.
6. Your inclination to buy varies by time of day. For retail websites the peak is 8 pm, for dating late at night, for finance around 1 pm, for travel just after 10 am.
7. Your e-mail address reveals your level of commitment.
8. Banner ads affect you more than you think. Advertising exerts a subconscious effect.
9. Companies win by not prompting customers to think.
10. Your web browsing reveals yur intentions.
11. Friends stick to the same cell phone company (a social effect).
12. Low credit rating, more car accidents.
13. Your shopping habits foretell your reliability as a debtor.
14. Small businesses' credit risk depends on the owner's behavior as a consumer.
15. Within a certain genetics cluster, having more genes shared by heterosexual couple means more infidelity by the female. We're programmed to avoid inbreeding, since there are benefits to genetic diversity.
16. Retirement is bad for your health, unhealthy habits such as smoking and drinking follow retirement.
17. Google search (Google Flu Trends) tends to predict disease outbreak. People with symptoms or in the vicinity of others with symptoms seek further information.
18. Smokers suffer less from repetitive motion disorder, because smokers take more breaks.
19. Positive health habits are contagious. People are strongly influenced by their social environment.
20. Happiness is contagious (a social effect). Each additional Facebook friend who is happy increases your chances of being happy by roughly 9%. Waves of happiness spread throughout the network.
21. Music expedites poststroke recovery and improves mood. Music listening activates a widespread bilateral network of brain regions related to attention, semantic processing, memory, motor functions and emotional processing.
22. Yoga improves your mood. Yoga is designed for, and practiced with the intent for, the attainment of tranquility.
23. Suicide bombers do not buy life insurance.
24. Unlike lighting, crime strikes twice.
25 Crime rises with public sporting events.
26. Crime rises after elections.
27. Music taste predicts political affiliations.
28. A job promotion can lead to quitting.
29. Vegetarians miss fewer flights.
30. Solo rockers die younger than those in bands.
Serendipity and Innovation
If necessity is the mother of invention, serendipity is its daddy. It was only by happy accident that Alexander Fleming happened upon the potent effects of penicillin, by noticing that an old bacterial culture he was about to clean up happened to be contaminated with some mold -- which was successfully killing it. By its very design, PA fosters serendipity. Predictive modeling conducts a broad, exploratory analysis, testing many predictors, and in so doing uncovers surprising findings, such as vegetarians being less likely to miss flights and so on.
The Ensemble Effect
(Netflix, CrowdSourcing)
Movie recommendations:
1. What's predicted: what rating a customer would give to a movie
2. What's done about it: customers are recommended movies that they are predicted to rate highly
The collective intelligence of a crowd emerges on many occasions:
1. Prediction markets, wherein a group of people together estimate the prospects for a horse race, political event, or economic occurrence by way of placing bets.
2. The audience of TV quiz show Who Wants to Be A Millionaire, whom contestants may poll to weigh in on questions.
3. Google's PageRank method, by which a web page's value and importance are informed by how many links people have created to point to the page.
4. The predictive capacity of the mass mood expressed by bloggers at large to foresee stock market behavior.
Human minds aren't the only things that can be effectively merged together. It turns out the aggregate effect emerging from a group extends also to nonhuman crowds -- of predictive models.
Like a crowd of people, an ensemble of predictive models benefits from the same "collective intelligence" effect. A Bag of Models = to predict, each model make its prediction and tally up the results. The Generalization Paradox: More is Less. When joined in an ensemble, predictive models compensate for one another's limitations, so the ensemble as w hole is more likley to predict correctly than its component models are.
A New Thing To Predict
Targeted Marketing with Response Modeling
1. What's predicted: which customers will purchase if contacted
2. What's done about it: contact those customers who are more likely to do so.
Comments
Post a Comment