Stock Market Prediction
NeuNet Pro BackProp Tutorial
"Stock Market Prediction"
- Using the NeuNet Pro FILE menu, select Create New BackProp Project.
Name this project "Stocks2".
- You are delivered to the Project Configuration screen.
Using the Data Source Browse button, select "STOCKS.NNS" as the data source for this project.
This database contains only one table "StockMarkets" so it will automatically be selected for you.
- The target field to be predicted is "Dow_Industrials".
Select this field in the list of available fields and pull it leftward into the Prediction Field by using the < button.
- Use the > button to move the following fields leftward into Input Fields.
These fields will be used to make the prediction.
- 10 Year
- 1 Year
- 90 Day
- Press the Advanced button and have a look at the Advanced Configuration screen.
It is not necessay to change anything on this screen.
Confirm the number of BackProp nodes is set to 5.
Press OK to return to Project Configuration.
- Complete the Project Configuration by pressing OK.
You are delivered to the Data Split screen.
- Split your data so rows 1 to 200 are used for training set and all 250 rows are used for testing set.
Simply drag the split handles up and down as necessary.
Hint: By double clicking on the number box, you can type in the split point.
Press OK when your split is complete.
You are delivered to the BackProp Training screen.
- Press GO to begin a continuous run of training cycles.
The prediction error will quickly decrease, as shown by the history graph.
Notice how the screen is constantly updated as the training proceeds.
Experiment with Graph Zoom.
Experiment with the STOP and GO buttons.
Try increasing Learn Rate and Momentum to 75 and notice how the history becomes choppy.
Notice how the blue coloring is used to flag which training cycle was saved as the best thus far.
That Best Cycle is always saved as the current version of your neural network.
Now reduce Momentum to 10 and Learn Rate to 20 and perform additional training cycles.
Experiment with Jog Weights.
- Press Stop, then press Finish.
A test cycle is automatically performed on your testing set,
and you are delivered to the Scatter Graph to see the results.
- The clustering of data points around the blue diagonal line shows your project is highly predictable.
Try double clicking on one of the poorly predicted points on the scatter graph:
You are delievered to browse all test data sorted by closeness to the point where you double clicked.
- Try the Time Series Graph and experiment with right-click and left-click on the Zoom button.
Remember the predictions left of week 200 were part of the training set,
while the predictions right of week 200 were not seen during training.
- Now have a look at Browse All and Browse Test Results.
Notice how you can right-click on any column header to sort and find within that column.
Notice how double-click or spacebar can be used to set a checkmark for any row.
- The prediction anomalies are your stock market buy and sell signals.
You can easily pull these anomalies to the top of the Browse Test Results table by performing a descending sort on the Difference column.
- Try using the Print button to print some of the graphs and tables.
- Congratulations on completing this tutorial.
For additional practice, go back to Configure This Project and try selecting some other fields,
then repeat the BackProp training process.
A Complete Neural Network Development System
CorMac Technologies Inc.
34 North Cumberland Street ~ Thunder Bay ON P7A 4L3 ~ Canada
E m a i l