What is Predictive Analytics?
As the name suggests, Predictive Analytics makes predictions about unknown future events. Data mining, statistics, modelling, machine learning and Artificial Intelligence can all play a role in analysing old and new data to make future predictions.
Patterns found in data can be used to identify risks and opportunities. Predictive Analytics models capture these relationships among many factors to assess risk with a particular set of conditions to assign a score or weight. By successfully applying Predictive Analytics, the organisation can effectively interpret big data for their benefit.
Is this a new phenomenon?
Predictive Analytics has been around for a long time. We have been calling it statistics since at least the beginning of the 1900’s. The major difference has a lot to do with the advances in technology. We can do a lot more than we could even 10 years ago for a few reasons:
We can store more data
We can collect more data due to the internet of things.
Computers are fast enough to analyse big data in an acceptable time-frame.
Why is predictive analytics important?
It provides an opportunity to identify future trends. This allows organisations to act on them ahead of time, thus increasing their bottom line and competitive advantage.
What is the difference between predictive analytics, descriptive analytics and prescriptive analytics?
- Predictive analytics is about understanding the future.
- Descriptive analytics is about insights into the past
- Prescriptive analytics offers advice on possible outcomes.
Give me an example of predictive analytics?
The most obvious example is in financial forecasting where predictive analytics can be applied to predict the future price of a commodity. Take for example gas. We know how much gas has cost over the last 100 years, and we know what external factors contribute to price changes. Using an analytics algorithm, historical pricing data and possible scenarios for today and tomorrow are analysed to predict future prices. The accuracy of the prediction weighs heavily on the quality of the algorithm, as well as the quality of the data.
Could it help me predict when I get to work?
Let’s say I want to predict what time you will get to work on Monday. You have provided me with a year’s data. This includes what time you got to work every Monday, how much sleep you got on Sunday nights, if you had meetings on Monday mornings, and if you went out of town for the weekend. I can make increasingly sophisticated predictions based on this data:
- The most naive prediction I can make is that you’ll get to work at the same time you did last week
- I can predict you’ll get there the average of the times you usually got there on Monday for the last year.
- I can propose a model that says the time you arrive is a function of some or all the information I listed above based on a year’s data:
Arrival time = f (sleep, meeting, weekend)