Eric Siegel answers eight questions about predictive analytics



my name is Erich Segal and my book is called predictive analytics predictive analytics is well the shortest definition is the subtitle of my book the power to predict who will click buy lie or die there's two kinds of people who care about predictive analytics number one all the organizations that win by making per person predictions this is companies governments hospitals universities law enforcement nonprofit even presidential campaigns these organizations win by predicting for each person to drive decisions as far as who to call who to send mail to who lend money to who to investigate for crime or fraud who to treat in certain ways for healthcare by doing this the organizations win they decrease risk Healthcare is made more robust they toughen crime-fighting sales are boosted for presidential campaigns more votes are gained number two the other kind of people who care about predictive analytics is everyone else because these organizations are making predictions on you and I every day we're being predicted as far as whether we're likely to buy whether we're going to heal get better with health care whether it will drop out of school whether we're going to commit theft we're going to steal something or commit a crime whether we're going to crash our car so all these predictions millions of predictions a day are being made about us by these organizations larger to our benefit definitely to their benefit you don't need to predict accurately to get great value organizations by predicting better than guessing get a little bit of you that between through the fog you know that blocks between today and tomorrow so by making these predictions per person that are a good bit better than guessing they actually played the numbers game better all organizations essential are playing numbers games and the way to optimize the vast operational scale of organizations today is with prediction three men alex is a big data thing well you know big data is basically just a grammatically incorrect way to say a lot of data but yeah the more data the better the more experience data essentially is experience of an organization's the aggregate or collective experience and the more you have the more there is from which to learn to create predictive models that make the per person individual predictions which will in turn be more accurate or more precise having been trained or learned over a greater amount of data so with today's excitement over big data it's all big data this big data that the question that begs is right well what's the point what are you going to do with it what's the most valuable you can think you can do with this greater and greater amount of data and the answer is the most actionable thing that organization can get out of data is learning from it how to make predictions per person because those per person predictions drive all the individual purpose and actions and decisions that organizations make no Nate silver didn't use predictive analytics to forecast the election however Obama did so silver made forecasts for overall state overall how with the prediction go whereas the Obama campaign actually use predictive analytics to make per voter decisions so that's really the difference between forecasting and predictive analytics predictive analytics makes it possible not just to predict the future but to influence it by driving these per individual person decisions by these individual predictions in the case of Obama's campaign the analytics group made these predictions to drive campaign decisions so while Nate Silver competed to predict the outcome of the election the Obama campaign competed to win the election itself and the thing that's interesting is instead of just predicting you know how each individual would vote are they more likely to vote for Obama or Romney are they likely to vote at all it's something completely different the Obama campaign predicted how likely is that this voter would be persuaded can we change their mind can we convince them will they be receptive to campaign contact a phone call a knock on the door and so by driving decisions with that way what they call predict persuasion modeling also known as uplift modeling they were actually able to gain more votes within individual swing states towards winning the election predictive analytics is powerful in that it produces these insights these predictions per each individual in some cases yes the thing that's being ascertained about the individual is very sensitive you know in the case of retail is that consumer pregnant in the case of large corporations is this employee likely to quit their job and the case of law enforcement is this incarcerated convict likely to commit a crime again if they're released so I like to quote spider-man's wise uncle who said with great power comes great responsibility because you know predictive analytics nobody would care if it weren't potent these things go together the fact it's so powerful that it brings up sensitive insights in some cases privacy in some cases to other civil liberties issues but there are some tricky things there's a lot that needs to be looked at and considered as as a society in terms of what to do with this newfound power and how to harness it in a safe way well as far as the underlying technology there are a lot of improvements taking place in predictive analytics and let me go over a couple of them right now number one uplift modeling that's predicting persuasion that's driving decisions as far as what's going to make the biggest difference this is what the Obama campaign uses I mentioned a moment ago they make decisions per voter which treatment which compact campaign contact or lack thereof is the best choice for each voter as far as swaying than in the right direction and avoiding the adverse effect of swaying them in the wrong direction very much the same thing with marketing I'm trying to sell something how am I going to persuade that person in health care which treatment medical treatment or lack thereof leaves a better chance of the positive outcome for that individual same core technology instead of predicting what a person is going to do the action or the outcome or behavior you're predicting will this treatment towards that person make a difference in the right direction that would like things to go that's called uplift modeling another hot trend in predictive analytics is what's called ensemble models so it turns out just like the collective intelligence over a crowd called the wisdom of the crowds where a bunch of people might come together and essentially vote or each put in their opinion and then the overall average opinion turns out to be better than most individual people you get the same effect with predictive models so the mechanisms that make predictions can in some case be pretty mundane that can be pretty primitive but when you group them together and there's not a lot of science or math you basically just pull them all together and make them vote just like people the models are voting and then the overall prediction machine is suddenly much better than most individual models so it's a great way to sort of tweak the robustness and correctness the precision the accuracy of how well these predictive models work simply by grouping them together and have pooling them so that they're basically what's called an ensemble mall well there's a lot of exciting really inspirational things going on in predictive analytics one of them is the IBM computer Watson that was able to successfully compete against the all-time human champions on the TV quiz show Jeopardy where those questions could be about anything they're intended for humans to answer the questions are very complex and they're grammar the written in English human language it turns out the predictive modeling the core analytical process of predictive analytics is the way Watson succeeds in choosing the answer it predicts is this candidate answer the correct answer to this question and in fact the way it does it so precisely is using what I mentioned a moment ago and ensemble models with using lots of models all together and it learns from historical previously given questions on you know from this TV shows history and is able to compete get one answer after another thing it questions could be about any topic the answer could be anything that picks out the one singular correct answer it's just it's just incredible you can see it on YouTube the actual broadcasted episodes of that TV show Jeopardy is just amazing it just rings off one correct answer after another you

8 Comments

  1. Matt L said:

    Another problem with Predictive Analytics when applied to business is meeting the statistical assumptions. How could you meet the normality assumption with business data? Most business data is very skewed. You could use non-parametric methods but that is weak. 

    June 27, 2019
    Reply
  2. Matt L said:

    Predictive Analytics and 'Big Data' is simply a business buzzword of this decade, you will probably never hear about this again in a couple of years because companies will find out that it does not work. It does not work in business because you cannot obtain data for all of the independent factors that has a casual relationship with the dependent factor, for example if your dependent is sales, how are you going to obtain data for factors such as brand equity, word of mouth sales, creative etc that might have a casual relationship with sales? What will happen when you do predictive analysis is an overcompensation for the independent factors that you have included in your model (see Omitted Variable Bias). Also, with business data, most of the independent factors you believe will have a casual relationship with the dependent factor will likely to show no relationship when you have conducted a sig test. If say the relationship between the factors are strong, how can you be sure that there is no confounding factor involved? Another fault with predictive is the bias modelling software they use (like eviews) which transforms data in a way that it fits into the model. It is like fitting a square into a circle. Because these software do not look at whether these factors should even be in a model, it is up to the user to determine what goes into the model, thereby there would be a possibility of bias. Business people that believes this should read up on 'Cause and Effect', Confounding factors, Omitted Variable Bias, Statistical Errors, Multicollinearity etc

    June 27, 2019
    Reply
  3. Sanjeev Robinson Paul said:

    nice video really informative..

    June 27, 2019
    Reply
  4. Perica Šandrk said:

    bullshit

    June 27, 2019
    Reply
  5. Philips Population Health Management said:

    It's interesting what this could have as an effect on patients into the future.

    June 27, 2019
    Reply
  6. John Gusiff said:

    If you aren't making Predictions and determining when you are right or wrong you have no means of Learning; and, therefore, be able to make better Business Decisions in the future.

    June 27, 2019
    Reply
  7. simeon24 said:

    *cough* prism, riot, perfect citizen, private interest *cough*

    June 27, 2019
    Reply
  8. Diane Lang said:

    Great video.. make more .. more .. please.

    June 27, 2019
    Reply

Leave a Reply

Your email address will not be published. Required fields are marked *