Analytics Builder > Working with Predictive Models
Working with Predictive Models
The Models List
The Analytics Builder provides a Models list page where all of the existing models for every available dataset are listed. The list includes the following information:
Column Name
Model Name
The name of a specific model. The name must be unique.
Dataset Name
The name of the dataset the model is based on.
List of the variables in a dataset that can serve as goals, (also known as dependent variables).
A filter that was applied to the dataset when the model was created. All models must have at least the all_data filter applied.
The current status of the model. When you first create the model, the state is "Queued" and then "Running." When the new model is created and ready for use, the state changes to "Completed." If the job fails, the state will show "Failed."
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.
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. For CONTINUOUS and ORDINAL goals.
RMSE Normalized
RMSE values normalized between 0 and 1.
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). For CONTINUOUS and ORDINAL goals.
Matthews Correlation Coefficient is a measurement of the quality of binary classifications. It takes into account the distribution of true and false positives in comparison with the distribution of true and false negatives. For BOOLEAN or CATEGORICAL goals.
A statistical measure, expressed in a percentage, that indicates how well the model predicted both successful and failed outcomes. For BOOLEAN and CATEGORICAL goals.
Number of Records in Validation Set
The number of records used to create the validation set. This value is based on the Validation Holdout % selected when the model was created.
Create Date
The date the model was created.
Model List Options
To create a new model, click New and follow the steps in Creating New Predictive Models.
To work with an existing model, select it in the models list and use one of the following options:
Delete – Removes the selected model.
View – Opens the Model Results page with a detailed view of the model statistics.
Job Details – Displays run time information for the model job.
Retrain – Generates a new model using the same configuration as the selected model. A prompt will request a new model name. The old model is retained along with any associated scoring jobs.
Publish – Publishes the completed model to Analytics Manager and opens it there as an analysis model
Export – Downloads the completed model as a PMML file. Only one model can be downloaded at a time.
Prev/Next – Pages back or forward through the list of models when the list is too long to display all at once.
Filtering the Model List
To filter the list of models displayed in the chart, click Add Filter and select from the list of columns (example: "Goals"). Then identify the conditions for filtering the column (example: Contains "Pump"). Click Save (only datasets with "Pump" included in the Goals column will be displayed).
Was this helpful?