Using Rapid Miner on Dynamic and Fast Trends Factors

astory

Dear all

I am looking into solutions on how to improvised my prediction model.

From my daily observations, the prediction model should only take into account records which happen earlier to predict future records.

E.g. 26/01/2016 predictions can only take in records before 25/01/2016. This is because 26/01/2016 records are usually highly correlated.

How can I use Rapid Miner to change the models and setup to cater to sequential modeling and validation ?

From my observations,
1. The accuracy of the Models against time is like Stock Market, Up and Down at repetitive Support and Resistance points.
   How can I change the data records or setup to cater into such conditions? Because at Support levels, prediction accuracy improve greatly and at resistance levels, accuracy certainly decrease drastically.

2. The nature of the problem is market behaviour dependent, the market may change trend to Follow trend or Averaging at times. However, it is repetitive every 2 - 4 days. There is no fix period though.
   How can I change the data records or setup to cater into such conditions?

3. Once in a blue moon (Around once in 2 months), there will be a period of erratic behaviours which there is a drastic Trend Change.
   -> For 2 - 3 days the accuracy for certain prediction may goes to 10% and below because of market changes, during which what data is available for the past 2-3 weeks are redundant and act as noise to prediction.
   -> During this period, the negate of predictions can achieve accuracy 90%. However, this is opposite from the main model.
   How can I change the data records or setup to cater into such exceptions?

Many thanks and Have a great week ahead!


Warmest Regards

Astor

JEdward

Have you downloaded the series extension yet?  It sounds like this is what you are looking for. 

Windowing will set the prediction horizon so that only earlier data is used for the prediction.
Classify by Trend changes the data to Up or Down (based on the trend).
You can add several Moving Averages to your dataset and use multiple in your predictions.