Optimising the performance of telecommunication bulk export using a machine learning closed loop system based on historic performance

Failures in telecommunication systems are typically resolved by the application of software patches or updates. These solutions tend to be specific to the failure type. Such bespoke system alterations are time consuming and financially expensive to implement. This paper proposes and evaluates a machine learning closed loop system to optimize the performance of the bulk configuration management data export. An evaluation file export service is developed and managed based on the industry standard JSR352 specification. The service produces failures reducing its overall performance. Rather than providing a specific solution for individual system failures, an adaptive and extensible machine learning closed loop system is introduced. The framework enables the file export service to learn from historic performance and to predict imminent failures. This type of closed loop system is optimized by providing the capability to re-learn based on new training data. This capability brings a dynamic approach to providing solutions. Solutions are reactive rather than static. Failures in software behavior can depend on environmental conditions like high load on a persistence layer, high volumes of traffic, insufficient hardware dimensioning. When failures occur due to factors like these, a dynamic redirection of the software to another flow stabilizes and improves the systems overall performance.