Machine Learning

Machine learning methods predict an outcome from a set of variables without assuming a specific mathematical form for how those variables relate to it — an alternative to Regression when the relationship between variables may be more complex than a linear model can capture.

Vesta offers two kinds of machine learning analysis, depending on the type of outcome you want to predict.

Machine Learning Regression

Use this when the outcome you want to predict is a continuous value. Results include each variable's importance to the prediction, alongside a plain-language summary explaining what the results mean.

Machine Learning Classification

Use this when the outcome you want to predict is binary (yes/no). Vesta offers eight classifiers to choose from. Results include variable importance, coefficients, and a ROC curve for every time step in a time-varying dataset.

Settings

Machine learning settings can be tuned to fit your data and question:

  • A slider controls how much of your data is held back for validation rather than used to fit the model.
  • Each algorithm has its own configurable options.
  • The report explains what each algorithm does, alongside its results.

Saving and Reusing a Model

A trained model can be saved and applied elsewhere, the same as with Regression and Logistic Regression models (see Regression).

Process Steps

  1. Click on the "Analyze" button from the side bar menu.
  2. Select the Machine Learning method appropriate to your outcome type — Regression for a continuous outcome, or Classification for a yes/no outcome.
  3. Choose Automatic, Guided, or Full control guidance, depending on how much control you want over the algorithm's settings. See Analysis Introduction for more on the Guided Analyze wizard.
  4. Select the dataset and variables to include.
  5. Select "Run" to fit the model.
  6. Results, including variable importance and a plain-language summary, are saved in the Data panel to the right.