Learning of a feature based on betti sequences obtained from time series data
An apparatus obtains time series data to which a label indicating a category of classification is attached, where the time series data contains variations of target values during a period to which the label is attached. The apparatus divides the time series data into a plurality of sections, and generates an attractor for each of the plurality of sections where the attractor indicates a set of numerical values toward which the variations of the target values tends to evolve. The apparatus calculates a Betti sequence in accordance with the generated attractor for each of the plurality of sections, and obtains a feature for the Betti sequences of the plurality of sections by clustering the Betti sequences or by calculating a statistical quantity of the Betti sequences. The apparatus performs machine learning by using the obtained feature as an input.
1. A learning method executed by a computer, the learning method comprising:
performing a feature extraction process including:
obtaining time series data to which a label indicating a category of classification is attached, the time series data containing variations of target values during a period to which the label is attached,
dividing the time series data into a plurality of sections, and generating an attractor for each of the plurality of sections, the attractor indicating a set of numerical values toward which the variations of the target values tends to evolve,
calculating a Betti sequence indicating a sequence of Betti numbers in accordance with the generated attractor for each of the plurality of sections, and
obtaining a feature for the Betti sequences of the plurality of sections by clustering the Betti sequences or calculating a statistical quantity of the Betti sequences; and
performing machine learning by using the obtained feature as an input.
2. The learning method of claim 1 , wherein
a feature vector is generated as the feature by using a number of sections belonging to each cluster used in the clustering the Betti sequences, and the machine learning is performed by using the feature vector as an input.
3. The learning method of claim 1 , wherein
at least one of mean, variance, kurtosis, and skewness is calculated as the statistical quantity of the Betti sequences, and the machine learning is performed by using the calculated statistical quantity as an input.
4. The learning method of claim 1 , wherein:
the time series data contains the variation of the target values obtained from a wearable sensor worn by an operator;
the feature extraction process is performed for each of unit times to obtain features for the respective unit times; and
the performing the machine learning includes training a classification model for classifying operation processes of the operator by using the obtained features.
5. The learning method of claim 4 , further comprising:
obtaining classification results by inputting the obtained features to the trained classification model; and
generating a daily operation report of the operator in accordance with the obtained classification results.
6. A non-transitory, computer-readable recording medium having stored therein a program for causing a computer to execute a process, the process comprising:
obtaining time series data to which a label indicating a category of classification is attached, the time series data containing variations of target values during a period to which the label is attached;
sectioning the time series data into a plurality of sections, and generating an attractor for each of the plurality of sections, the attractor indicating a set of numerical values toward which the variations of the target values tends to evolve;
calculating a Betti sequence indicating a sequence of Betti numbers in accordance with the generated attractor for each of the plurality of sections;
obtaining a feature for the Betti sequences of the plurality of sections by clustering the Betti sequences or by calculating a statistical quantity of the Betti sequences; and
performing machine learning by using the obtained feature as an input.
7. An apparatus comprising:
a memory; and
a processor coupled to the memory and configured to:
obtain time series data to which a label indicating a category of classification is attached, the time series data containing variations of target values during a period to which the label is attached,
divide the time series data into a plurality of sections, and generate an attractor for each of the plurality of sections, the attractor indicating a set of numerical values toward which the variations of the target values tends to evolve,
calculate a Betti sequence indicating a sequence of Betti numbers in accordance with the generated attractor for each of the plurality of sections,
obtain a feature for the Betti sequences of the plurality of sections by clustering the Betti sequences or by calculating a statistical quantity of the Betti sequences, and
perform machine learning by using the obtained feature as an input.