Creating an Anomaly Detection Model

Anomaly detection is very useful for detecting outliers in your data. For example, anomaly detection is ideal for detecting fraudulent credit card transactions or detecting cyber attacks. When training an anomaly detection model the training data should have no anomalies in it, that way, when it is trained all of the variables can be compared to find a standard range in which all "normal" data exists in. With Terrene you want to select all of the variables that might affect each other (this means you should leave out identifiers such as name, email, card numbers, etc).

Creating a Workspace

To start we must first create a new workspace (or add to an existing workspace, the steps are nearly identical, more information can be found here) by clicking "Create a New Workspace".

Dashboard

Now you will be given the option to name your new workspace and give it a brief description.

Name Workspace

Now you can click "Continue" and we will upload our data.

Uploading Your Data

Next you have to upload the dataset you will be training the anomaly detection model on. For more information on connecting a database click here or uploading a CSV click here.

Variable Selection

Once you have successfully uploaded a dataset you will be able to select the variables you are interested in comparing. First, you must select the Anomaly Detection tab since Terrene defaults to General Purpose Predictive Model.

Selection Anomaly Inputs

Now you may select your input variables. It is important to note, there are no output variables for anomaly detection. When a prediction is made the model will take a new set of inputs and tell you whether or not that piece of data was an anomaly. Once you click "Finish" the model will begin training and you will be taken to the workspace dashboard where you can see the training progress and make predictions.