Predictive analytics is a methodology based on "machine learning" that makes it possible to anticipate which students are at risk. It can be accessed through Applications > Predictive models:

In the upper-right corner, the Create model button will appear. From there, you will be asked to provide the name, description, data source, and algorithm (neural network or linear regression).

The data source must include:

• The basic model data (name, data source, etc.).
• The case identifiers: e.g., an ID field that identifies the student, such as Student ID, national ID number, or email address.
• The prediction outcome: e.g., whether the student will pass or fail the course. It must have a value of 0 or 1.
• The metrics you want to consider when predicting the outcome: for example, connection time, number of connections, access to forums, completed assignments, published posts, grades, etc.
• Select a column containing zeros and ones, where 0 indicates that the row contains historical data for training the model, and 1 means that the row will be subject to prediction. You can also use the value -1 if you want to ignore the values in that row.

Once the model has been created, if you click it, you can view its parameters, train it, and check its validity by clicking it again. 

For the purposes of validating the predictions, Precision and Specificity are particularly relevant and should be at least 0.8. In the graph, the farther the blue curve is from the red dashed line, the more reliable the prediction will be.