Learning system, analysis system, learning method, and storage medium
A learning system including: a training data acquisition unit that, for each of a plurality of feature amounts obtained by converting time-series data on a predetermined period basis, acquires training data with which an event type in the predetermined period is associated; and a classifier learning unit that performs learning on a plurality of classifiers different from each other by using a feature amount group including one or more feature amounts corresponding to some or all periods out of the plurality of feature amounts included in the training data to perform learning on a classifier, which classifies the event type, for multiple times while changing a combination in the feature amount group, and the event type includes a first event type corresponding to a state where a target event to be classified occurs and a second event type corresponding to a state where the target event does not occur.
1. A learning system comprising:
at least one memory storing instructions; and
at least one processor configured to execute instructions to:
acquire, for each of a plurality of feature amounts obtained by converting time-series data on a predetermined period basis, training data with which an event type in the predetermined period is associated; and
perform learning on a plurality of classifiers that are different from each other by using a feature amount group including one or more feature amounts corresponding to some or all of periods out of the plurality of feature amounts included in the training data to perform an operation of performing learning on a classifier, which classifies the event type, for multiple times while changing a combination in the feature amount group,
wherein the event type includes a first event type corresponding to a state where a target event to be classified is occurring and a second event type corresponding to a state where the target event is not occurring,
wherein out of the feature amount group used for learning on the classifier, the number of feature amounts associated with the second event type is larger than the number of feature amounts associated with the first event type,
wherein the learning is performed by using all of the feature amounts corresponding to the first event type out of the plurality of feature amounts and using some of the feature amounts corresponding to the second event type out of the plurality of feature amounts, and
wherein the plurality of feature amounts include at least a variance of the time-series data in the predetermined period.
2. The learning system according to claim 1 , wherein the feature amount group used for learning on the classifier is selected at random.
3. The learning system according to claim 1 , wherein the plurality of feature amounts include a feature amount based on time-series data corresponding to only at least one of occurrence time and end time of the target event.
4. The learning system according to claim 1 , wherein the time-series data is partitioned into periods which equally divide a day.
5. The learning system according to claim 1 , wherein the learning is performed by using a combination of feature amounts in which all of feature amounts based on continuous time-series data for at least one day are included with respect to feature amounts corresponding to the second event type out of the plurality of feature amount.
6. An analysis system comprising:
the learning system according to claim 1 ; and
a plurality of classifiers, each trained by the learning system configured to perform classification of an event type corresponding to a plurality of feature amounts in a classification target period.
7. The analysis system according to claim 6 , wherein the plurality of classifiers determine that an event corresponding to an event type classified by the most classifiers occurred in the classification target period.
8. The analysis system according to claim 6 ,
wherein the plurality of classifiers are configured to:
determine that an event corresponding to an event type classified by the most classifiers occurred in the classification target period when the maximum number of classifiers that have performed classification into each event type is larger than a predetermined threshold, and
determine that the event does not occur in the classification target period when the maximum number of classifiers that have performed classification into each event type is smaller than or equal to the predetermined threshold.
9. The learning system according to claim 1 , wherein out of the feature amount group, the feature amounts associated with the second event type used for learning on the classifier is selected at random by using a randomized algorithm.
10. The learning system according to claim 1 , wherein out of the feature amount group, the feature amounts associated with the second event type used for learning on the classifier is selected at random by using a randomized algorithm.
11. A learning method comprising:
for each of a plurality of feature amounts obtained by converting time-series data on a predetermined period basis, acquiring training data with which an event type in the predetermined period is associated; and
performing learning on a plurality of classifiers that are different from each other by using a feature amount group including one or more feature amounts corresponding to some or all of periods out of the plurality of feature amounts included in the training data to perform an operation of performing learning on a classifier, which classifies the event type, for multiple times while changing a combination in the feature amount group,
wherein the event type includes a first event type corresponding to a state where a target event to be classified is occurring and a second event type corresponding to a state where the target event is not occurring,
wherein out of the feature amount group used for learning on the classifier, the number of feature amounts associated with the second event type is larger than the number of feature amounts associated with the first event type,
wherein the learning is performed by using all of the feature amounts corresponding to the first event type out of the plurality of feature amounts and using some of the feature amounts corresponding to the second event type out of the plurality of feature amounts, and
wherein the plurality of feature amounts include at least a variance of the time-series data in the predetermined period.
12. A non-transitory storage medium storing a program that causes a computer to perform a learning method comprising:
for each of a plurality of feature amounts obtained by converting time-series data on a predetermined period basis, acquiring training data with which an event type in the predetermined period is associated; and
performing learning on a plurality of classifiers that are different from each other by using a feature amount group including one or more feature amounts corresponding to some or all of periods out of the plurality of feature amounts included in the training data to perform an operation of performing learning on a classifier, which classifies the event type, for multiple times while changing a combination in the feature amount group,
wherein the event type includes a first event type corresponding to a state where a target event to be classified is occurring and a second event type corresponding to a state where the target event is not occurring, and
wherein out of the feature amount group used for learning on the classifier, the number of feature amounts associated with the second event type is larger than the number of feature amounts associated with the first event type,
wherein the learning is performed by using all of the feature amounts corresponding to the first event type out of the plurality of feature amounts and using some of the feature amounts corresponding to the second event type out of the plurality of feature amounts, and
wherein the plurality of feature amounts include at least a variance of the time-series data in the predetermined period.