Classification and Prediction of Business Incidents Using Deep Learning for Anomaly Detection
Name
Samreen Mahak Hassan
Abstract
Companies today use a number of software systems to carry out various business activities. Such enterprise standard software solutions consist of a large number of components usually developed by different teams and/or different software vendors using various technologies. In such complex software systems, there can be various issues ranging from problems in the software itself to issues in network.
In order to measure the operational performance of applications and infrastructure as well as key performance indicators (KPIs) that evaluate the success of the organization, a lot of business metrics is collected. These metrics have certain data patterns which represent normal business behaviour. Anomalies are some unexpected changes within these data patterns such as degradation or a sudden surge in business metrics values. Additionally, small changes in software system configuration can cause unexpected behaviour in business flows. Version upgrades of different components can introduce compatibility problems. These problems could lead to a change in the normal behaviour of business metrics and cause anomalies. These anomalies if not resolved quickly results in business and financial losses. Therefore, it is necessary for businesses to take proactive steps to manage such business incidents before they can adversely affect it. This brings us to the need for an analytics platform which can analyze patterns of data streams, identify and differentiate normal behaviour of a business metric from anomalous behaviour and could generate a notification.
The current anomaly detection and alert system in Playtech plc uses a simple anomaly detection technique that follows a rule based approach and it is observed that it is not efficient. Thus, a more robust, modular and efficient business incident/anomaly detection solution based on advanced machine learning techniques is needed that could work in conjugation with the current system. This thesis proposes, describes and evaluates a business incident/anomaly detection system based on deep learning approach that categorises and predicts the business incidents/anomalies using the available business metrics information.
In order to measure the operational performance of applications and infrastructure as well as key performance indicators (KPIs) that evaluate the success of the organization, a lot of business metrics is collected. These metrics have certain data patterns which represent normal business behaviour. Anomalies are some unexpected changes within these data patterns such as degradation or a sudden surge in business metrics values. Additionally, small changes in software system configuration can cause unexpected behaviour in business flows. Version upgrades of different components can introduce compatibility problems. These problems could lead to a change in the normal behaviour of business metrics and cause anomalies. These anomalies if not resolved quickly results in business and financial losses. Therefore, it is necessary for businesses to take proactive steps to manage such business incidents before they can adversely affect it. This brings us to the need for an analytics platform which can analyze patterns of data streams, identify and differentiate normal behaviour of a business metric from anomalous behaviour and could generate a notification.
The current anomaly detection and alert system in Playtech plc uses a simple anomaly detection technique that follows a rule based approach and it is observed that it is not efficient. Thus, a more robust, modular and efficient business incident/anomaly detection solution based on advanced machine learning techniques is needed that could work in conjugation with the current system. This thesis proposes, describes and evaluates a business incident/anomaly detection system based on deep learning approach that categorises and predicts the business incidents/anomalies using the available business metrics information.
Graduation Thesis language
English
Graduation Thesis type
Master - Computer Science
Supervisor(s)
Aleksandr Tavgen, Martin Kiilo and Raimundas Matulevičius
Defence year
2019