The concept responsible for predictive analytics methods is to use mathematical algorithms to locate routines and general trends in data. From those patterns and trends, details are gleaned that will help business managers appreciate causal elements that may be pushing behavior through their agency. What's more, the models created by predictive analytics tools could be used to "score" or "predict" a set of data to ascertain others which may go with exactly the same behavior pattern down the line.
And so the essential principle is, a business person selects 1 of 2 things:
1. A subject he or she would like to compare with the other group.
2. A numeric field that person desires to forecast.
Let's have a look at each of these two different instances:
A good example of comparing a target population may be to examine human resources information around corporate pay. One example is, consider indicating that data by using a scatterplot where the horizontal axis is overall performance (from low toward the left to high on the right) and the vertical axis is pay (from low on the bottom to high toward the top). One would anticipate finding a growing line up from the lower left towards the upper right - - (e.g., those staff members with more substantial performance need to be paid much more, while individuals with very low performance really need to be paid less.) Invariably, however, you will see outliers particularly the top left quadrant which happen to be high paid personnel with very low overall performance ratings. This could be a target. So to execute predictive analytics tools, the operator would pick the top left quadrant of very high paid/low employees, and then use a very simple point-and-click interface to run a regression-based model that should do a comparison of that target to the whole entire populace of all employees. The output would be a model describing which factors most explain the outlying behavior of the target. That model could then be used to score or rank the remainder of the people to name employees that are "at risk" to potentially have the similar behavior the future of high pay relative to performance This style of model is called a "classification model".
And so the essential principle is, a business person selects 1 of 2 things:
1. A subject he or she would like to compare with the other group.
2. A numeric field that person desires to forecast.
Let's have a look at each of these two different instances:
A good example of comparing a target population may be to examine human resources information around corporate pay. One example is, consider indicating that data by using a scatterplot where the horizontal axis is overall performance (from low toward the left to high on the right) and the vertical axis is pay (from low on the bottom to high toward the top). One would anticipate finding a growing line up from the lower left towards the upper right - - (e.g., those staff members with more substantial performance need to be paid much more, while individuals with very low performance really need to be paid less.) Invariably, however, you will see outliers particularly the top left quadrant which happen to be high paid personnel with very low overall performance ratings. This could be a target. So to execute predictive analytics tools, the operator would pick the top left quadrant of very high paid/low employees, and then use a very simple point-and-click interface to run a regression-based model that should do a comparison of that target to the whole entire populace of all employees. The output would be a model describing which factors most explain the outlying behavior of the target. That model could then be used to score or rank the remainder of the people to name employees that are "at risk" to potentially have the similar behavior the future of high pay relative to performance This style of model is called a "classification model".
About the Author:
Learn more about Predictive Analytics Tools by visiting Predictive Analytics Webinars and Tutorials