Data Science – Anomaly Detection

Anomaly Detection module checks your data for anomalies and notifies you of any inconsistencies in your time-series data. This is extremely useful to monitor:

  • Sales
  • Website Visitors
  • Users Actions (purchases)
  • Quantities of transactions

Any kind of data that is important to your business can be monitored to give you the peace of mind that you will be the first to know.

Step 1 – Select a Table

Start by choosing the table or view that has the data that you would like to monitor. In our case bellow we are adding some monitoring on a table that has FX rates.

Step 2 – Select Date column

A date column is the one that defines the timely flow of your data. Usually it is the timestamp of when the records was created, such as created_at field.

Step 3 – Select Value column

Next up is to choose the value column of the data that you would like to be analysed for anomalies. Here you can choose to count, average or sum the values before the anomaly algorithm is executed. Here we would like to take average rate (fx rate)

Step 4 – Add filters (optional)

Here you can choose to filter your data to run anomaly only on the subset of your data, such as a particular product or service. Use the Query Preview window to see exactly the query we will run to fetch the data for Anomaly analysis.

Step 5 – Choose the lookback interval

The parameters forecast days and lookback interval can be left as they are. Lookback interval is the only important parameter for Anomaly Detection because it determines how far back the data is gathered for the model. Smaller lookback period will build a model that will be reflective of the recent trends, while a larger lookback window will build a model that will be based on longer-term trend.

Once you are satisfied with your model, click Run to test it and save it.

Tags: anomaly, data science, forecast

Was this helpful?