Anomaly Detection is a vast area of data analytics. The approach to Anomaly Detection depends on which problem area/ Business use case you are focusing on. Anomaly detection has a different meaning with respect to different Industries or domains.
What is Time Series Anomaly Detection?
As the name suggests, Time Series Anomaly Detection deals with Time Series data, i.e. Data which changes with time. E.g. In our personal computer, CPU usage, Network Usage, Memory Usage with time. You can plot it as a line graph with time in the X-axis. It clearly shows the behaviour of the metric over a period of time.
Time series Anomaly Detection involves applying AI/ML algorithms on the time series data to understand the behaviour of the data and then call out any abnormal behaviour such as sudden spike or dip.
What are various types of anomalies depending on the scenarios?
1. Point Anomalies:
If data instance can be observed against other data instances such as an anomaly, then it is considered as point anomaly.
2. Contextual Anomalies:
In this case, the data instance is anomalous in some predefined context.
It designates a group of instances that exhibits similar anomalous behavior compared to other groups.
Why Time Series Anomaly Detection?
It is Vertical Agnostic. If you pick up any business, they will have various metrics or KPIs to monitor the performance of their business. This helps to get a sense of how good or bad their performance is. The data for these metrics will follow a time-series pattern, which can be used for Time Series Anomaly Detection. Since all the metrics follow time, we can use the time as a common feature to tie various similar behaving metrics together by applying correlation which can help the business to focus on the incident with the list of all impacted metrics. It will be very helpful for finding out the root cause in case of an incident.
What should the Anomaly detection tool do for you?
- Connect to all the different Data Sources where the metrics are generated.
- Use Proprietary unsupervised algorithms that are created for understanding and learn all the trend, periodicity and seasonality in the data.
- In Real-time, it should identify all emerging issues autonomously which deviates from the normal behavior.
- Various anomalies can be grouped using correlation which will be useful for doing root cause analysis in case of any incident.
- Provide Smart insights which can be consumed by Business users without any dependency on Data Analysts.
- Reduce significant efforts in configuration and development efforts so that the Network Company can focus on the results and take meaningful decisions.
How can Anomaly Detection Tool add any Value?
At CrunchMetrics, we use ensemble models through proprietary algorithms to give you high-level accuracy and performance tweaked for any Business scenarios in any verticals or industries. The tool does the whole process of Data ingestion from various sources, Training of data, Anomaly Detection, Anomaly Scoring, Correlation, Alerts, and Notifications to various stakeholders for you.
Reach out to us to learn more about our Anomaly Detection Software
Kshitish Sahoo is the product manager at CrunchMetrics. He has more than 7 years of experience in domains such as Energy and Utilities, QSR, Healthcare, CPG, Real Estate, Banking and Insurance and E-Commerce. He has worked on setting the strategy, developing the feature propositions, marketing the product and handling the financial metrics of the product.