What Exactly Is An Anomaly?
Anomaly detection algorithms define the concept of an expected outcome that deviates from the standard. In data analysis, we frequently stumble upon anomalies when we try to observe a dataset and notice that the expected pattern is not kept.
An unexplainable drop in sales, the Corona Virus outbreak, losing your credit card data, or anything similar are great anomaly detection examples.
Times Series and Anomaly Detection
Time series data consists of observations made over a defined period of time. The data set provided will have a time stamp. In time series, the anomaly detection algorithm will be based on observing historical data that doesn’t conform to what was to be expected. This is what Google has implemented within their analytics.
Non-Time Series Anomaly Detection
Anomaly detection data doesn’t necessarily have to be applied in a time series. You could also be looking for anomalies that aren’t consistent and do not fall into the expected data clusters.
There are tons of ways to categorize anomalies due to their unpredictability. However, we will focus on the explanation provided by DataScience.com that narrows anomaly detection classification down to the three most important:
- Point anomalies: A specific instance of data that is nowhere near any other historical data. This is applicable for credit cards when an abnormal amount is removed from a said credit card. Another anomaly could be considered receiving an online casino free bonus no deposit. And if you follow the link, it won’t even be an anomaly.
- Contextual anomalies: Abnormality that is context-specific. Mostly used for time-series data, when purchasing USD100 worth of food would normally be suspicions. When done during the holidays, it becomes easily explainable.
- Collective anomalies: Collecting a set of data instances in order to detect an anomaly. This can be used in business and applies when someone is trying to send data to a server from various different locations simultaneously. This can sometimes be the inception of a cyber-attack.
Alerting for Google Analytics
Ever since Google launched its Analytics software, traffico anomalo Google in Spanish, you could set up alerts. The previous alerts were quite static, and you could not properly customize them. Their required maintenance proved to be quite the issue. For example, a 20% increase or decrease in traffic during an entire month wouldn’t be very relevant while you are building your business. If your business was already developed, then it would be highly valuable information.
If you use absolute values instead, the information would be even more useless. If your alert is set up at 50 new visits during business infancy, it would be highly relevant. For a matured business having new visits becomes irrelevant. If you already had these alerts enabled, you probably know that the false positive frequency is insanely high. Those days are over because of the introduction of Anomaly Detection.
The New Feature – Traffico Anomalo Google
Machine learning algorithms will detect anything that might be relevant for your business. Google also introduced an alert when anomaly detection applications are detected. You will receive a graph that highlights the anomalies so that you can take a closer look and decide if they are relevant for your business.
Google identifies what is expected according to the historical anomaly detection data; they call this “forecasted value.” Whenever something doesn’t fall within expected values, they send a notification. They take a look at historic data to create the upper and lower limits using the Bayesian space-time model. They also make sure to ensure that the detections have statistical validity. An important take-away is the fact that Google only uses time series anomaly data detection, and there is no way to control what metric or dimension they will monitor.
How Can you Put Google’s Anomaly Detection to Good Use?
The alerts you are going to receive can lead to valuable conclusions that will greatly aid your business. If you have a large increase or decrease in page views, you can immediately investigate and see what caused it. If it is an increase, you can try to replicate it. If it’s a decrease, you can try to avoid it in the future. Anomaly detection Google will be an amazing resource.
If you were still using the static alerts that Google was offering previously, there would be a huge chance that you would miss a relevant increase or decrease. This would mainly be because you wouldn’t have a proper alert setup enabled without Google Analytics Anomaly Detection.
We hope that Google will further improve the functionality of anomaly detection methods. Any advancement in the software can prove to be a huge aid regardless of the state your business is in. We hope that they will allow us to prioritize certain metrics in the future, and you can find uses for them to grow your business. We see the great potential behind Google Anomaly detection, and it should help most of you tackle issues you didn’t even know you had.
Have you already been putting Google Analytics alerts to good use? Share your findings in the comments and see what our other readers have been doing with the software.