MACHINE LEARNING FOR PREVENTIVE ASSURANCE AND RECOVERY ACTION OPTIMIZATION

摘要

#$%^&*AU2019202683A120190509.pdf#####ABSTRACT A computer-implemented method executed by one or more processors, the method including receiving behavior data and line parameter data from a plurality of user devices in real-time, associated with a plurality of communication lines, each user device associated with a respective communication line, wherein the behavior data includes information regarding bandwidth usage and utilization time frames, and the line parameter data includes information regarding the respective communication line and devices that access the respective communication line including device availability, line availability, boot times, link retrains, bit rates upstream and downstream, central processing unit load and signal to noise margin ratio, training a predictive model using training data to determine correlations, wherein the training data is received behavior data and received line parameter data for a subset of the plurality of user devices and a subset of the plurality of communication lines for a particular temporal interval, and wherein at least one correlation indicates that the probability of a network anomaly grows with an increasing number of reboots, line drops or bitrate upstream/downstream, and wherein at least one correlation indicates that the probability of a network anomaly decreases when central processing unit load, signal to noise margin ratio or line availability increases, processing received behavior data and line parameter data through the predictive model, repeatedly tuning the predictive model, providing a risk score for each communication line of the plurality of communication lines based on the processing, each risk score representing a likelihood that a network anomaly for the respective communication line would occur within a determined temporal period, and selectively performing test recovery actions for a communication line based on a respective risk score, the one or more recovery actions performed to inhibit occurrence of network anomalies, wherein selectively performing the test recovery actions includes dividing communication lines from the plurality of communication lines into subsets and applying different test recovery actions to different subsets of communications lines, and measuring an effect that each test recovery action has on a respective risk score for respective communication lines, wherein the communication lines are ordered according to the respective risk scores, and the recovery actions are selectively performed automatically based upon the respective risk score meeting a determined threshold.1/4 00 a)w