Artificial intelligence-based latency anomaly detection in a wireless communications system

1 . A system comprising:

at least one hardware processor; and

at least one non-transitory memory storing instructions, which, when executed by the at least one hardware processor, cause the system to:

obtain traffic data served by a user plane function of a core network;

determine, using a first machine learning model, multiple traffic patterns using the traffic data served by the user plane function,

wherein the multiple traffic patterns correspond to one or more anomalies in a network latency of the core network, and

wherein the first machine learning model is configured using multiple training patterns that correspond to an expected network latency of the core network;

select a second machine learning model from multiple machine learning models using a respective query runtime of each of the multiple machine learning models;

determine, using the second machine learning model, one or more user plane function events causing the one or more anomalies,

wherein the one or more user plane function events is determined using the traffic data, and

wherein the second machine learning model is configured to:

identify a portion of the traffic data that includes the one or more anomalies, and

provide the one or more user plane function events using the portion of the traffic data; and

transmit a reconfiguration command to the core network to adjust one or more routing policies of the user plane function using the determined one or more user plane function events, wherein the reconfiguration command is configured to align the network latency with the multiple training patterns.

2 . The system of claim 1 , wherein the system is caused to:

form a training dataset by applying at least one of data cleaning, feature engineering, or data transformations to the traffic data; and

train at least one of the first machine learning model or the second machine learning model using the training dataset.

3 . The system of claim 1 ,

wherein the one or more anomalies is a first set of anomalies,

wherein the first machine learning model is configured to generate a second set of anomalies using a change of the network latency, and

wherein the change of the network latency includes changes in one or more of: average latency, peak latency, or latency distribution over time.

4 . The system of claim 1 ,

wherein the system is caused to measure the respective query runtime of each of the multiple machine learning models to process the traffic data, and

wherein the measurement of the respective query runtime is based on one or more of: execution time, memory usage, or computational efficiency of a corresponding model.

5 . The system of claim 1 , wherein the second machine learning model has a lower query runtime than the respective query runtime of each of the multiple machine learning models.

6 . The system of claim 1 ,

wherein the system is caused to trigger one or more alarms, and

wherein the one or more user plane function events satisfy one or more predetermined criteria.

7 . The system of claim 1 ,

wherein at least one of: the first machine learning model or the second machine learning model is one or more of: an anomaly detection model, a forecasting model, or a trend detection model,

wherein the anomaly detection model is configured to identify one or more outliers within the traffic data,

wherein the forecasting model is configured to predict one or more future trends associated with the network latency within the core network based on the traffic data and the one or more outliers, and

wherein the trend detection model is configured to identify one or more patterns that indicate expected traffic data using the one or more future trends, the one or more outliers, and historical traffic data.

8 . A non-transitory, computer-readable storage medium comprising instructions recorded thereon, wherein the instructions when executed by at least one data processor of a computer system, cause the computer system to:

obtain traffic data served by a user plane function of a core network;

determine, using a first machine learning model, multiple traffic patterns using the traffic data served by the user plane function,

wherein the multiple traffic patterns correspond to one or more anomalies in a network latency of the core network, and

wherein the first machine learning model is configured using multiple training patterns that correspond to an expected network latency of the core network;

determine, using a second machine learning model, one or more user plane function events causing the one or more anomalies,

wherein the one or more user plane function events is determined using the traffic data, and

wherein the second machine learning model is configured to:

identify a portion of the traffic data that includes the one or more anomalies, and

provide the one or more user plane function events using the portion of the traffic data; and

transmit a reconfiguration command to the core network to adjust one or more routing policies of the user plane function using the determined one or more user plane function events, wherein the reconfiguration command is configured to align the network latency with the multiple training patterns.

9 . The non-transitory, computer-readable storage medium of claim 8 , wherein the instructions cause the computer system to:

form a training dataset by applying at least one of data cleaning, feature engineering, or data transformations to the traffic data; and

train at least one of the first machine learning model or the second machine learning model using the training dataset.

10 . The non-transitory, computer-readable storage medium of claim 8 ,

wherein the one or more anomalies is a first set of anomalies,

wherein the first machine learning model is configured to generate a second set of anomalies using a change of the network latency, and

wherein the change of the network latency includes changes in one or more of: average latency, peak latency, or latency distribution over time.

11 . The non-transitory, computer-readable storage medium of claim 8 ,

wherein the instructions cause the computer system to measure a respective query runtime of each of multiple machine learning models to process the traffic data, and

wherein the measurement of the respective query runtime is based on one or more of: execution time, memory usage, or computational efficiency of a corresponding model.

12 . The non-transitory, computer-readable storage medium of claim 8 , wherein the second machine learning model has a lower query runtime than a respective query runtime of each of multiple machine learning models.

13 . The non-transitory, computer-readable storage medium of claim 8 ,

wherein the instructions cause the computer system to trigger one or more alarms, and

wherein the one or more user plane function events satisfy one or more predetermined criteria.

14 . The non-transitory, computer-readable storage medium of claim 8 ,

wherein at least one of: the first machine learning model or the second machine learning model is one or more of: an anomaly detection model, a forecasting model, or a trend detection model,

wherein the anomaly detection model is configured to identify one or more outliers within the traffic data,

wherein the forecasting model is configured to predict one or more future trends associated with the network latency within the core network based on the traffic data and the one or more outliers, and

wherein the trend detection model is configured to identify one or more patterns that indicate expected traffic data using the one or more future trends, the one or more outliers, and historical traffic data.

15 . A method comprising:

obtaining traffic data served by a user plane function of a telecommunications network;

determining, using a first machine learning model, a set of traffic patterns using the traffic data served by the user plane function,

wherein one or more traffic patterns in the set of traffic patterns correspond to one or more anomalies in a network latency of the telecommunications network, and

wherein the first machine learning model is configured using training patterns that correspond to an expected network latency of the telecommunications network;

determining, using a second machine learning model, one or more user plane function events causing the one or more anomalies,

wherein the one or more user plane function events is determined using the traffic data, and

wherein the second machine learning model is configured to:

identify a portion of the traffic data that includes the one or more anomalies, and

provide the one or more user plane function events using the portion of the traffic data; and

transmitting a reconfiguration command to the telecommunications network to adjust one or more routing policies of the user plane function using the determined one or more user plane function events, wherein the reconfiguration command is configured to align the network latency with the set of training patterns.

16 . The method of claim 15 , the method comprising:

forming a training dataset by applying at least one of data cleaning, feature engineering, or data transformations to the traffic data; and

training at least one of the first machine learning model or the second machine learning model using the training dataset.

17 . The method of claim 15 ,

wherein the one or more anomalies is a first set of anomalies,

wherein the first machine learning model is configured to generate a second set of anomalies using a change of the network latency, and

wherein the change of the network latency includes changes in one or more of: average latency, peak latency, or latency distribution over time.

18 . The method of claim 15 , the method comprising measuring a query runtime of each model in multiple machine learning models to process the traffic data,

wherein the measurement of the query runtime is based on at least one of: execution time, memory usage, or computational efficiency of a corresponding mode.

19 . The method of claim 15 , wherein the second machine learning model has a lower query runtime than a respective query runtime of each of multiple machine learning models.

20 . The method of claim 15 ,

wherein at least one of: the first machine learning model or the second machine learning model is one or more of: an anomaly detection model, a forecasting model, or a trend detection model,

wherein the anomaly detection model is configured to identify one or more outliers within the traffic data,

wherein the forecasting model is configured to predict one or more future trends associated with the network latency within the core network based on the traffic data and the one or more outliers, and

wherein the trend detection model is configured to identify one or more patterns that indicate expected traffic data using the one or more future trends, the one or more outliers, and historical traffic data.