# Becoming Fluent in Predictive Analytics

November 26, 2013 7:34 AM Subscribe

I work in a University managing the broad based direct mail, email and calling programs. I have zero undergrad or graduate experience with math, business or the social sciences. (Aka, I can write a really nice essay...) I would like to chart a path to being recognized as an expert in predictive analytics.

I am willing to dedicate up to two years to my professional development in this area. I have free access to my University's business and math departments. I assume I'll need some kind of basic math foundation before jumping into statistics? I haven't taken any kind of math class in 15 years and I hardly remember how to solve for X. I will need a course of education that will start with the basics and works up to using SPSS. Thank you!

I am willing to dedicate up to two years to my professional development in this area. I have free access to my University's business and math departments. I assume I'll need some kind of basic math foundation before jumping into statistics? I haven't taken any kind of math class in 15 years and I hardly remember how to solve for X. I will need a course of education that will start with the basics and works up to using SPSS. Thank you!

Best answer: On the math side, you need a basic foundation in linear algebra and calculus. You need to be able to understand how people approach and solve optimization problems (e.g., how do you find the value of x that makes the function f(x) as large as possible?) You need to understand what vectors and matrices are.

Once you have this foundation, you can turn to probability and statistics. A basic statistical tool is linear regression, which basically involves fitting a straight line so that it matches the data as closely as possible (this is where optimization comes in; also, your data is most conveniently represented in vector/matrix form). There are a number of other statistical tools that can be viewed as generalizations of linear regression -- logistic regression, generalized linear models, and so on. It is also helpful to have a basic understanding of probability distributions (normal distribution, Poisson, exponential, etc) -- statistical modelers often assume that certain aspects of the data follow certain probability distributions.

Generally speaking, running a statistical model using software is not that difficult -- you can often find an online guide for any particular model. The hard part is interpreting the output and figuring out which models should be run and which is best. This requires experience and judgment, but it also requires a solid foundation in the basics. At the same time, there is a learning curve for using any bit of statistical software -- you need to put in the hours and work on real projects with it.

Another hard part of the predictive analytics process is pre-modeling -- that is, in accumulating and organizing and cleaning the data that is eventually modeled. This is a practical skill that is generally best learned "on the job" with real projects. Familiarity with a programming language like Python may help, as will a working knowledge of SQL (a language for manipulating database information). At the same time, predictive analytics projects often involve teams, so it's not like one person needs to be expert in every part of the process.

The business side of things is technically easier, but it involves knowing which questions to ask -- which lines of inquiry are likely to be useful and fruitful, and which are likely to be unproductive? I'm not sure this can be successfully generalized either -- what is important for one business may be unrelated to another business.

I would be surprised if starting from scratch you would be recognized as an "expert" within two years, but you can certainly make a good start. I know people with solid math/statistics/business backgrounds who struggle with predictive analytics because they have mental blocks around some of the material. In general, I think this is an area in which people dedicated to lifelong learning tend to do well.

posted by leopard at 8:05 AM on November 26, 2013 [11 favorites]

Once you have this foundation, you can turn to probability and statistics. A basic statistical tool is linear regression, which basically involves fitting a straight line so that it matches the data as closely as possible (this is where optimization comes in; also, your data is most conveniently represented in vector/matrix form). There are a number of other statistical tools that can be viewed as generalizations of linear regression -- logistic regression, generalized linear models, and so on. It is also helpful to have a basic understanding of probability distributions (normal distribution, Poisson, exponential, etc) -- statistical modelers often assume that certain aspects of the data follow certain probability distributions.

Generally speaking, running a statistical model using software is not that difficult -- you can often find an online guide for any particular model. The hard part is interpreting the output and figuring out which models should be run and which is best. This requires experience and judgment, but it also requires a solid foundation in the basics. At the same time, there is a learning curve for using any bit of statistical software -- you need to put in the hours and work on real projects with it.

Another hard part of the predictive analytics process is pre-modeling -- that is, in accumulating and organizing and cleaning the data that is eventually modeled. This is a practical skill that is generally best learned "on the job" with real projects. Familiarity with a programming language like Python may help, as will a working knowledge of SQL (a language for manipulating database information). At the same time, predictive analytics projects often involve teams, so it's not like one person needs to be expert in every part of the process.

The business side of things is technically easier, but it involves knowing which questions to ask -- which lines of inquiry are likely to be useful and fruitful, and which are likely to be unproductive? I'm not sure this can be successfully generalized either -- what is important for one business may be unrelated to another business.

I would be surprised if starting from scratch you would be recognized as an "expert" within two years, but you can certainly make a good start. I know people with solid math/statistics/business backgrounds who struggle with predictive analytics because they have mental blocks around some of the material. In general, I think this is an area in which people dedicated to lifelong learning tend to do well.

posted by leopard at 8:05 AM on November 26, 2013 [11 favorites]

Response by poster: Leopard - this was AWESOME. Exactly what I needed. Thank you for the thorough and thoughtful answer!

posted by meta x zen at 11:36 AM on November 26, 2013

posted by meta x zen at 11:36 AM on November 26, 2013

« Older Is there a temporary aesthetic solution to... | Do you deal with major depression without drugs? Newer »

This thread is closed to new comments.

posted by evoque at 7:45 AM on November 26, 2013