Analytics Builder > Working with Predictive Models > View Model Results
  
View Model Results
Overview
Model results are displayed as a set of statistics followed by a graph which show how accurately the predictive model reflects real outcomes from the data. The specific statistics, and the type of graph displayed, both depend on the goal variable type. For more information, see the Model Result Statistics and Model Result Graphs sections below. In addition to the result graph and statistics, the Model Results page also provides buttons to view:
• Model Details — Opens a Model Statistics and Features dialog box that displays a set of detailed statistical properties (with their values) and a list of the features included in the model.
• Model Configuration — Displays the configuration used to generate the model, including key parameters and learner techniques.
To View the Results
1. On the Models list page, find and select the model for which you want to view results.
2. Click the View button.
The Model Results page opens.
Model Result Statistics
Above the result graph, a set of statistics is displayed for the selected model. Which statistics are displayed for a given model depends on the goal variable type. Possible statistics include the following:
Field Name
Description
Applies to Goal Types:
Model Status
The current state of the model. Can include Queued, Running, Completed, or Failed.
All
Model Result ID
An automatically generated identifier for the model results.
All
ROC
Represents the area under the ROC (Receiver Operating Characteristics) curve. This statistic indicates how well the model separates positives and negatives. It’s a measurement of the model's ability to correctly classify predictions as true or false across various discrimination thresholds. Displays when either a ROC curve or a confusion matrix graph is shown.
* 
If you retrain a model that was first generated in an earlier release of ThingWorx Analytics, the ROC value might be different in the retrained model. This change is the result of an enhancement, made in the 8.2 release, to the calculation of area under the ROC curve.
Boolean
Precision
Represents the fraction of instances, classified as positive, that are actually positive. This statistic indicates the exactness or quality of the results. A model has maximum precision if all of its positive predictions are correct (even if some positives are missed). A model has poor precision if it incorrectly classifies many negatives as positive. Displays when either a ROC curve or a confusion matrix graph is shown.
Precision and Recall are often considered together to interpret the quality and completeness of prediction results. A model can exhibit high precision but low recall, if it classifies everything as negative. Conversely, a model can exhibit low precision but high recall, if it classifies everything as positive.
Boolean
Recall
Represents the fraction of all positive instances that are correctly classified as positive. This statistic indicates the completeness of the results, or the extent to which true positives are not missed. A model has maximum recall if it correctly classifies all of the positive instances (even if it also incorrectly classifies all of the negatives as positive). A model has poor recall if it misses many positives (even if it correctly classifies all the negatives). Displays when either a ROC curve or a confusion matrix graph is shown.
Recall is also known as Sensitivity or True Positive Rate (TPR).
Boolean
Specificity
Represents the fraction of all negative instances that are correctly classified as negative. This statistic indicates how well the model avoided incorrectly classifying negative instances as positive (false positives). A model has perfect specificity if it correctly classifies all negative instances as negative (even if it incorrectly classifies all of the positives as negative). A model has poor specificity if it incorrectly classifies most negatives as positives. Displays when either a ROC curve or a confusion matrix graph is shown.
Specificity is also known as True Negative Rate (TNR).
Sensitivity (Recall) and Specificity are often considered together to interpret how often the model detects the result it is looking for, and how often the model mistakes something else for the result it is looking for. A model can exhibit high sensitivity but low specificity, if it classifies everything as positive. Conversely, a model can exhibit low sensitivity but high specificity, if it classifies everything as negative.
Boolean
RMSE
Root Mean Square Error is a measurement of the difference between values predicted by the model and the values actually observed. A low RMSE value is better than a high value. Displays when a bubble plot is shown.
Continuous
Pearson Correlation
A measure of the linear correlation (or dependence) between the predicted and actual results. Values can range between -1 (total negative correlation) and +1 (total positive correlation). A score of 0 shows no correlation at all. Displays when a bubble plot is shown.
Continuous
R-squared
Represents the fraction of variance in a goal (dependent variable) that is predicted by the independent variables in the model. This statistic indicates how well the model fits the data. The R-squared value will be 1 if all of the predictions were perfect. Displays when a bubble plot is shown.
Continuous
Adjusted R-squared
Similar to the R-squared value but adjusted to account for the number of variables included in the model. This calculation tries to account for the fact that R-squared accuracy increases whenever additional variables are added to the model, even if those variables are not relevant to predicting the goal. Adjusted R-squared can be useful for comparing models created with different numbers of variables. It allows selection of the model with the best fit, without unnecessary additional variables. Displays when a bubble plot is shown.
Continuous
Model Result Graphs
The type of result graph displayed for a specific model is based on the data type of the goal variable in the model. The following graph types are described in the sections below:
• ROC Curve
• Confusion Matrix
• Bubble Plot
ROC Curve
A ROC curve plots the true positive rates against false positive rates at various discrimination thresholds. For a good model, the curve will rise quickly on the left and stay close to the top, showing that the model predictions were accurate. Displayed only when the goal variable is a Boolean.
ROC Curve sample
* 
When a Boolean goal value is selected during model creation, the Results Graph offers the option to view either a ROC Curve or a Confusion Matrix. Both graphs are available on separate tabs that you can toggle between.
Confusion Matrix
A confusion matrix shows the number of correct and incorrect predictions the model produced, compared to the actual outcomes. Displayed only when the goal is a Boolean variable.
Confusion Matrix sample
Bubble Plot
A bubble plot graphs the approximate location of records along x,y axes, where the x-axis represents actual results and the y-axis represents predicted results. Displayed whenever the goal variable is Continuous. Because this data is binned for charting, each bubble on the graph represents a range of one or more records that have predicted and actual results near the bubble location. For each bubble, a tooltip displays the value at the center of the range for both actual and predicted results in the bubble.
The size of each bubble is approximated to represent the relative number of results at each location. Bubble size does not correlate to specific record counts. Bubbles of the same size do not necessarily represent the same number of records.
The bubble plot graph uses color to show whether the result was over-predicted, accurately-predicted, or under-predicted as follows:
highly over-predicted Highly over-predicted – Includes predicted results that are 75% above the actual value.
over-predicted Over-predicted – Includes predicted results that are 25 to 75% over the actual value.
accurate Accurate – Includes predicted results that are within 25% of the actual value.
under-predicted Under-predicted – Includes predicted results that are 25 to 75% below the actual value.
highly under-predicted Highly under-predicted – Includes predicted results that are 75% below the actual value.
Bubble Plot sample
* 
The predicted and actual results are displayed in the tooltips as double values, rather than integers. These results are binned for charting and the data points on the chart do not represent individual records. Each point represents a range and the tooltip displays the predicted and actual values at the center of that range.