Machine Learning: An introduction to Predictive science

One question that has always been on my mind is this

Is our behavior completely random that we cannot develop some sort of pattern and predict future behavior?

I disagree that there isn’t a pattern to behavior, there is a pattern to almost anything and everything and can be translated to some sort of scientific algorithm. I use “almost” because I do agree there are exceptions but I am satisfied with the 66% predictibity.

So we get into statistics, which by the way I am a statistics geek, having spent some time in my career in predictive science. Lets discuss trend analysis

What is a trend? “a line of general direction or movement”. We must establish something, what is the general direction of the data? Is it upwards, downwards or in-between? Through statistic we can establish the general trend of the data and based on this predict the next data. Again, based on the “trend”. Take a look at this graph

This shows a downward trend so we expect the prediction for 6 will be below 1000 but it will not below the 5 value of 600. This is because we factor in the average line which will pass through 800+ and so the figure for 6 will be close to that average line.

Now when we get the actual figure for 6 we will examine the actual compared to the predicted and the computer will learn from that deviation how to adjust its prediction for 7, gradually the computer will learn the data trend from past trends and future deviations and improve in its prediction. That is machine learning in a high level 101.

I advise you get a series of data like a 10 yr history, plug in a 5 year history then use the predictive system to predict for year 6 and then use your actual data for 6 to measure deviation and then continue like that up to year 10, which will gives the system the ability to learn from history.

I will be posting a predictive system soon and using some sample data (like monthly grocery bills) to show you how predict future behavior. Then in another post I will show you how to use this system to predict the behavior of Microsoft Identity System (MIM 2016) in Operations. We will be able to predict

  • When will users change their passwords?
  • How do users change their passwords?
  • When do users change their address?
  • When will their be the next bulk sync?