Databases Reference
In-Depth Information
Drag SelectedHomes down over the ANN regression model and release.
Select “Generate predictions”.
When a dataset is dropped on a model, if VisMiner finds all model input
attributes in the dropped dataset, it will use those inputs to generate predicted
output values for all rows, then create a new dataset containing the original data
plus a new attribute having the same name as the output attribute followed by the
word “Predicted”. In this case the new name is “pricePredicted”.
Change the name of this new dataset to an easier “predictedPrice”.
Right-click on the dataset; select “Create derived dataset”.
“Select All” existing attributes.
Create a new column named pcntGain with the formula: (pricePredicted
price)/price
100.
Name the dataset “gains”.
Click “Create”.
Open gains in a parallel coordinate plot.
Drag the bottom slider of the priceGain column up until only 50 observa-
tions are visible (or as close to 50 as the display precision will allow you to
get).
Save this dataset as “top50”.
You now have a dataset containing the 50 best bargains according to the
model built to predict price.
Summary
Regression analysis is used to build models that generate predictions for a
continuous numeric output attribute.
VisMiner implements three regression algorithms: linear regression, polyno-
mial regression, and artificial neural networks (ANN). Of the three, the ANN
algorithm is the only one that can detect interaction between input attributes.
In comparing alternative regression models, a common measure of model
performance is R
2
, which reflects the percentage of variation in output values
explained by the model. To assess the contributions that input attributes make
toward a predicted output value, VisMiner implements regression model surface
plots. The surface plots are also very useful
in detecting and evaluating
interaction effects between inputs.
