The purpose of anomaly detection is to facilitate the work of professionals detecting unusual or emergency/fraud events from a large amount of data.
Most of the time these processes are either not automated at all or only one source system can be examined at a time. The purpose of machine learning supported anomaly detection is to find correlations between coherent or seemingly independent systems, and odd events that, based on their parameters, do not fit into the operating process, differ from normal operation. Events found by the AI can be examined in detail by a human expert or pre-defined automatisms.
Main features
- Ability to collect large and complex datasets (e.g.: log data)
- Creating custom tables (fully customized type, shape, etc. of the input logs)
- Finding anomalies using AI
- Notifying operators
- User friendly and interactive visual interface
Success criteria
- Sufficient teaching data set
- It is suggested to collect data from multiple locations within the organization so that the most complex errors can be detected
- The source systems from which data was collected for teaching must provide data in real time
- It is necessary to fine-tune / re-teach the model periodically for even more accurate results.
Advantages
- In most cases, the necessary data is available to the company but they are not collected properly.
- This solution can speed up bug fixing processes, furthermore after proper pre-training they can also be alerted before occurring. .