Appropriate algorithm for time series prediction of an event


  • Hello

    here's what I have: a number of quantities that vary in time; let's say I have N different integers each 5 minutes. From time to time, an event E happens.

    I would like to know, by monitoring the values of the N integers in time, when the event E is about to happen again.

    I expect an answer like "when Quantity 1 goes over 110 and then Quantity 2 drops 30% in less than 20 minutes, then in another 10 minutes E has a 90% chance to happen".

    What should I do? Which algorithm should I use (if there is such an algorithm)?

    I am sorry i couldn't find other answers in the msdn so i am asking it here. Feel free to point me to other threads if necessary.

    Thank You for Your help.


    Sunday, September 08, 2013 9:23 PM


All replies

  • Hi Wentu,

    Microsoft Time Series algorithm provides regression algorithms that are optimized for the forecasting of continuous values, such as product sales, over time. Please take a look at the following article:

    Microsoft Time Series Algorithm:

    In addition, here is an article regarding create Time Series Predictions for Mining model in SQL Server Analysis Services. Please see:

    If I have something misunderstood, please point out and elaborate your requirement with more detail.

    Best Regads,

    Elvis Long
    TechNet Community Support

    Wednesday, September 11, 2013 6:09 AM
  • Hello Mr. Long and Thankc for Your reply

    from what I understand, in Time Series Prediction I have a quantity, let's say Q, and I know how Q varies with time. With these data I try to predict the value of Q itself in the near future.

    What I need is different. Let's say I have some quantities, let's say Q,R,S, and again I know their values in the past. Besides, I know that a certain event E happens from time to time (and I know the timestamps of these occurrencies). I want to know, starting from the present values of Q,R and S, if the event E is about to happen again in the near future. E is not a time series, it has not a value that i can track with time, E is more a collection of timestamps and i want to know which is the next timestamp that i will have. A classical example is hardware failure. I measure the temperature and the reads per second of a hard disk and i want to be alerted that the event E = hard disk breakdown is about to happen, based on the historical data of when other hard disks broke in relation to their temperatures just before the break down.

    Thank You very much


    Sunday, October 06, 2013 12:45 PM