Systems and methods for network incident management
Systems, apparatuses, and methods are provided herein for network incident management. A method for network incident management comprises aggregating network metrics associated with a monitored network in a historical network data database, identifying incidents based on the network metrics, generating a training data set based on the network metrics and the incidents, wherein the training data set comprises time series of network metrics as training input and incidents as labels, training an incident model using the training data set, receiving real-time network metrics from the network via the network interface, determining an incident prediction based on the incident model using the real-time network metrics as input, and causing a user interface device to provide an alert to a user based on the incident prediction.
1. A system for network incident management, the system comprises:
a historical network data database;
a network interface configured to receive data from a monitored network comprising a plurality of components; and
a control circuit coupled to the historical network data database and the network interface, the control circuit being configured to:
aggregate network metrics from the monitored network in the historical network data database, wherein the monitored network comprises an e-commerce server and the aggregated network metrics include e-commerce orders per minute (OPM), response time, and network latency corresponding to e-commerce channel(s) of the e-commerce server;
identify incidents based on the aggregated network metrics;
generate, based on the aggregated network metrics and the incidents, a training data set that includes multiple time series for the OPM, the response time, and the network latency as training input, wherein historical incidents are labeled on the multiple time series;
train, with a machine learning algorithm, an incident model using the training data set;
receive real-time network metrics from the monitored network via the network interface;
predict future incident(s) based on the trained incident model using the real-time network metrics as input; and
cause a user interface device to provide an alert to a user based on the predicted future incident(s).
2. The system of claim 1 , wherein the aggregated network metrics further comprise page views, throughput, network congestion signals, or HTTP codes.
3. The system of claim 1 , wherein the training data set is generated based on oversampling the aggregated network metrics and the predicted future incident(s).
4. The system of claim 1 , wherein the incident model is a long-term short memory model (LSTM) trained based on shifting time windows of the time series of network metrics and incidents.
5. The system of claim 1 , wherein the predicted future incident(s) comprises a future point in time and a likelihood indicator.
6. The system of claim 1 , wherein the predicted future incident(s) comprises an incident type indicator, an incident severity indicator, or an indication of a network component.
7. The system of claim 1 , wherein the control circuit is further configured to automatically send an instruction to at least one component of the monitored network to cause a change at the at least one component of the monitored network based on the predicted future incident(s) before a predicted incident time.
8. The system of claim 1 , wherein the control circuit is further configured to update the incident model based on the real-time network metrics and detected incidents.
9. The system of claim 1 , wherein the plurality of components comprises a central processing unit cluster, a storage cluster, and a database cluster.
10. A method for network incident management, the method comprises:
aggregating, by a control circuit and via a network interface, network metrics associated with a monitored network in a historical network data database, wherein the monitored network comprises an e-commerce server and the aggregated network metrics include e-commerce orders per minute (OPM), response time, and network latency corresponding to e-commerce channel(s) of the e-commerce server;
identifying, with the control circuit, incidents based on the aggregated network metrics;
generating, with the control circuit, based on the aggregated network metrics and the incidents, a training data set that includes multiple time series for the OPM, the response time, and the network latency as training input, wherein historical incidents are labeled on the multiple time series;
training, with a machine learning algorithm executed on the control circuit, an incident model using the training data set;
receiving real-time network metrics from the monitored network via the network interface;
predicting future incident(s) based on the trained incident model using the real-time network metrics as input; and
causing a user interface device to provide an alert to a user based on the predicted future incident(s).
11. The method of claim 10 , wherein the aggregated network metrics further comprise page views, throughput, network congestion signals, or HTTP codes.
12. The method of claim 10 , wherein the training data set is generated based on oversampling the aggregated network metrics and the predicted future incident(s).
13. The method of claim 10 , wherein the incident model is a long-term short memory model (LSTM) trained based on shifting time windows of the time series of network metrics and incidents.
14. The method of claim 10 , wherein the predicted future incident(s) comprises a future point in time and a likelihood indicator.
15. The method of claim 10 , wherein the predicted future incident(s) comprises an incident type indicator, an incident severity indicator, or an indication of a network component.
16. The method of claim 10 , further comprising:
automatically send an instruction to at least one component of the monitored network to cause a change at the at least one component of the monitored network based on the predicted future incident(s) before a predicted incident time.
17. The method of claim 10 , further comprising:
updating the incident model based on the real-time network metrics and detected incidents.
18. An apparatus comprising:
a processor; and
a non-transitory storage medium storing computer readable instructions that, when executed by the processor, cause the processor to:
aggregate network metrics from a monitored network in a historical network data database, wherein the monitored network comprises an e-commerce server and the aggregated network metrics include e-commerce orders per minute (OPM), response time, and network latency corresponding to e-commerce channel(s) of the e-commerce server;
identify incidents based on the aggregated network metrics;
generate, based on the aggregated network metrics and the incidents, a training data set that includes multiple time series for the OPM, the response time, and the network latency as training input, wherein historical incidents are labeled on the multiple time series;
train, with a machine learning algorithm, an incident model using the training data set;
receive real-time network metrics from the monitored network via a network interface;
predict future incident(s) based on the trained incident model using the real-time network metrics as input; and
cause a user interface device to provide an alert to a user based on the predicted future incident(s).