Traffic sensor, analysis method, and analysis program
A traffic sensor includes processing circuitry configured to calculate a degree of spread of a range of normal communication indicated by a normal communication model for each of the normal communication model for detecting abnormal communication of an Internet of Things (IoT) device learned for each of the IoT device to be monitored, classify a normal communication model in which the degree of spread is less than a predetermined value as a normal communication model of an IoT device of a first model, and classify a normal communication model in which the degree of spread is equal to or greater than the predetermined value as a normal communication model of an IoT device of a second model, detect abnormal communication in the IoT device by using the normal communication model of the IoT device of the first model, and extract a feature amount.
1 . A traffic sensor comprising:
processing circuitry configured to:
calculate a degree of spread of a range of normal communication indicated by a normal communication model for each of the normal communication model for detecting abnormal communication of an Internet of Things (IoT) device learned for each of the IoT device to be monitored;
classify a normal communication model in which the degree of spread is less than a predetermined value as a normal communication model of an IoT device of a first model, and classify a normal communication model in which the degree of spread is equal to or greater than the predetermined value as a normal communication model of an IoT device of a second model;
detect abnormal communication in the IoT device by using the normal communication model of the IoT device of the first model;
extract a feature amount in which a contribution degree to detection of the abnormal communication is equal to or greater than a predetermined value among feature amounts used in the normal communication model of the IoT device of the first model;
reconstruct the normal communication model of the IoT device of the second model using the feature amount that has been extracted;
analyze abnormal communication common to each of IoT devices by using abnormal communication of the IoT device of the first model detected using the normal communication model and abnormal communication of the IoT device of the second model detected using the normal communication model that has been reconstructed; and
calculate the degree of spread of the range of the normal communication indicated by the normal communication model by a product of variances in feature amount directions used for the normal communication model.
2 . The traffic sensor according to claim 1 ,
wherein the processing circuitry is further configured to output abnormal communication having a similarity equal to or greater than a predetermined value as a result of analysis of abnormal communication of each of the IoT devices.
3 . The traffic sensor according to claim 1 ,
wherein the processing circuitry is further configured to, when a connection of a new IoT device is detected, learn a normal communication model of the IoT device based on communication data of the IoT device.
4 . An analysis method performed by a traffic sensor, the analysis method comprising:
calculating a degree of spread of a range of normal communication indicated by a normal communication model for each of the normal communication model for detecting abnormal communication of an Internet of Things (IoT) device learned for each of the IoT device to be monitored;
classifying a normal communication model in which the degree of spread is less than a predetermined value as a normal communication model of an IoT device of a first model, and classifying a normal communication model in which the degree of spread is equal to or greater than the predetermined value as a normal communication model of an IoT device of a second model;
detecting abnormal communication in the IoT device by using the normal communication model of the IoT device of the first model;
extracting a feature amount in which a contribution degree to detection of the abnormal communication is equal to or greater than a predetermined value among feature amounts used in the normal communication model of the IoT device of the first model;
reconstructing the normal communication model of the IoT device of the second model using the feature amount that has been extracted;
analyzing abnormal communication common to each of IoT devices by using abnormal communication of the IoT device of the first model detected using the normal communication model and abnormal communication of the IoT device of the second model detected using the normal communication model that has been reconstructed; and
calculating the degree of spread of the range of the normal communication indicated by the normal communication model by a product of variances in feature amount directions used for the normal communication model.
5 . A non-transitory computer-readable recording medium storing therein an analysis program that causes a computer to execute a process comprising:
calculating a degree of spread of a range of normal communication indicated by a normal communication model for each of the normal communication model for detecting abnormal communication of an Internet of Things (IoT) device learned for each of the IoT device to be monitored;
classifying a normal communication model in which the degree of spread is less than a predetermined value as a normal communication model of an IoT device of a first model, and classifying a normal communication model in which the degree of spread is equal to or greater than the predetermined value as a normal communication model of an IoT device of a second model;
detecting abnormal communication in the IoT device by using the normal communication model of the IoT device of the first model;
extracting a feature amount in which a contribution degree to detection of the abnormal communication is equal to or greater than a predetermined value among feature amounts used in the normal communication model of the IoT device of the first model;
reconstructing the normal communication model of the IoT device of the second model using the feature amount that has been extracted;
analyzing abnormal communication common to each of IoT devices by using abnormal communication of the IoT device of the first model detected using the normal communication model and abnormal communication of the IoT device of the second model detected using the normal communication model that has been reconstructed; and
calculating the degree of spread of the range of the normal communication indicated by the normal communication model by a product of variances in feature amount directions used for the normal communication model.
6 . The analysis method according to claim 4 , further comprising:
outputting abnormal communication having a similarity equal to or greater than a predetermined value as a result of analysis of abnormal communication of each of the IoT devices.
7 . The analysis method according to claim 4 , further comprising:
when a connection of a new IoT device is detected, learning a normal communication model of the IoT device based on communication data of the IoT device.
8 . The non-transitory computer-readable recording medium according to claim 5 , further comprising;
outputting abnormal communication having a similarity equal to or greater than a predetermined value as a result of analysis of abnormal communication of each of the IoT devices.
9 . The non-transitory computer-readable recording medium according to claim 5 , further comprising:
when a connection of a new IoT device is detected, learning a normal communication model of the IoT device based on communication data of the IoT device.