Prediction Model Generation
What Is a Prediction Model?
ThingWorx Analytics ingests and analyzes data from connected devices. It uses a set of predictive analytic algorithms and machine learning techniques to identify meaningful patterns in the data. Based on this analysis of the existing data, a generalized prediction model is generated. This model can then be applied to subsequent data to make predictions associated with specific outcomes (such as machine failures, performance issues, risk of downtime).
How is a Prediction Model Generated?
Generating a prediction model in ThingWorx Analytics can involve both training the model and validating it. Training is the process by which ThingWorx Analytics uses a number of machine learning modeling techniques to construct a prediction model. The training process can hold out a percentage of the training data for validation purposes. If a validation holdout is specified, a separate process runs to validate the model against the validation set.
At the end of the prediction model training process, ThingWorx Analytics publishes a ready-for-use model, in PMML format, that you can use to start making predictions (also referred to as Scoring). If a validation process ran, it generates a separate set of results.
What Types of Data Can Be Predicted?
During training, the data type of the goal variable affects which learner techniques can be used and the amount of time required to generate a model. ThingWorx Analytics can analyze the following types of data:
• Continuous – Numeric data that can be represented on a continuous scale, such as temperature
• Boolean – Data that has only two values, such as true or false
• Categorical – Data that can be grouped with no numerical value or order among the groups, such as hair color or gender
• Ordinal – Data that can be grouped into ordered categories, such as economic status (low, medium, high) or ranking (first, second, third)
|
|
When using a categorical goal variable, the number of categories represented in your data can affect the performance of both model generation and predictive scoring procedures. A dataset with a large number of categories can increase processing time significantly.
|
For information about the modeling techniques available, and the ways in which they can be combined to generate accurate predictive models, see
Learners and Ensemble Techniques.
How to Access Predictive Model Functionality
ThingWorx Analytics predictive model generating functionality can be accessed via the following methods:
• ThingWorx API – In ThingWorx Composer, prediction model generation is handled through a services in a Training Thing. These services can be used to submit training requests, retrieve model results, list jobs, and more. Requires installation of both ThingWorx Foundation and ThingWorx Analytics Server .
• Analytics Builder – As part of the ThingWorx Analytics Extension, Analytics Builder provides a user interface for interacting with your data. In addition to loading and configuring a dataset in Analytics Builder, you can also generate a predictive model. Requires installation of both ThingWorx Foundation and ThingWorx Analytics Server.
