Working On A Time Series Tutorial

Posted on by

Its hard to believe but I'm already 50% done with my first batch of Rapidminer 5.0 Video Tutorials.  So far, the reception has been positive and I thank everyone who emailed me, IM'd me, or commented on these tutorials.  I made a promise to myself to make a batch of 10 video tutorials first before I re-engineer this website further.

So far so good and I'm on target to record a new video tutorial tonight or tomorrow morning.  Video #6 will be about creating decision trees in Rapidminer  for a direct mail marketing example.  For video #7, I'll probably focus on an evolutionary weighting example, and then close out the remaining 3 tutorials with financial time series examples.

Below is an screenshot of one of the time series examples I'm working on. This is a time series chart of the S&P500 with a neural net generated trend line.  Neat, huh?

  • Bob

    I'm very curious about the figure you posted. It appears to me that either 1) your model has absolutely amazing predicative capabilities and you will be very wealthy very soon, or 2) your model is using future information to estimate the past.

    At the time/index ~240 the close reaches it's lowest point, however the trend line is already sloping up suggesting amazing predictive accuracy or extreme data snooping bias.

    Please elaborate! 

  • http://www.neuralmarkettrends.com Tom

    @Bob: This is nothing to get excited about.  The ANN generated a trend curve based on the above data series set, its not forecasting anything.  The data is from the past and its nothing more than a nonlinear trend line based on that data set.  I'll elaborate more when I record the tutorial.