Reconstructing time series datasets with missing values utilizing machine learning
An apparatus comprises a processing device configured to obtain a time series dataset having missing values, the time series dataset comprising monitoring data associated with one or more assets. The processing device is also configured to generate, utilizing a machine learning algorithm, a reconstructed time series dataset having imputed values for the missing values in the obtained time series dataset, the machine learning algorithm comprising a generative network implementing inverse network parameter determination for network parameters of the generative network. The processing device is further configured to classify patterns in the obtained time series dataset utilizing the reconstructed time series dataset, to select remedial actions for controlling at least one of the one or more assets based at least in part on the classified patterns in the obtained time series dataset, and to initiate the selected remedial actions to control the at least one asset.
1. An apparatus comprising:
at least one processing device comprising a processor coupled to a memory;
the at least one processing device being configured to perform steps of:
obtaining a time series dataset having one or more missing values, the time series dataset comprising monitoring data associated with one or more assets;
generating, utilizing a machine learning algorithm, a reconstructed time series dataset having one or more imputed values for the one or more missing values in the obtained time series dataset, the machine learning algorithm comprising a generative network implementing inverse network parameter determination for network parameters of the generative network;
classifying one or more patterns in the obtained time series dataset utilizing the reconstructed time series dataset;
selecting one or more remedial actions for controlling at least one of the one or more assets based at least in part on the classified one or more patterns in the obtained time series dataset; and
initiating the selected one or more remedial actions to control the at least one asset;
wherein the machine learning algorithm comprises a convolutional neural network; and
wherein generating the reconstructed time series dataset having the one or more imputed values comprises performing two or more iterations of:
applying the obtained time series dataset having the one or more missing values to the convolutional neural network to produce a candidate reconstructed time series dataset having one or more imputed values for the one or more missing values in the obtained time series dataset; and
tuning network parameters of the convolutional neural network using a loss function that compares the candidate reconstructed time series dataset with the obtained time series dataset.
2. The apparatus of claim 1 wherein the time series dataset comprises a univariate time series dataset.
3. The apparatus of claim 1 wherein the convolutional neural network is initialized with randomized values for the network parameters.
4. The apparatus of claim 1 wherein the loss function comprises a sum of a variation loss and a regularization of the network parameters of the convolutional neural network.
5. The apparatus of claim 1 wherein the convolutional neural network comprises two or more layers each comprising two or more convolutional filters.
6. The apparatus of claim 1 wherein the convolutional neural network comprises:
at least a first layer that applies batch normalization and utilizes a rectified linear unit activation function;
at least a second layer that does not apply batch normalization and utilizes a hyperbolic tangent activation function; and
at least a third layer that comprises a fully connected layer.
7. The apparatus of claim 6 wherein the third layer comprises a final layer of the convolutional neural network, and wherein the convolutional neural network comprise two or more instances of the first layer and at least one instance of the second layer prior to the final layer.
8. The apparatus of claim 1 wherein classifying one or more patterns in the obtained time series dataset utilizing the reconstructed time series dataset comprises applying the reconstructed time series dataset as input to at least one additional machine learning algorithm.
9. The apparatus of claim 1 wherein the time series dataset comprises telemetry data associated with the one or more assets.
10. The apparatus of claim 9 wherein the one or more assets comprise one or more computing resources in an information technology infrastructure.
11. The apparatus of claim 10 wherein the classified one or more patterns in the obtained time series dataset characterize health of at least a given one of the one or more computing resources in the information technology infrastructure, and wherein the selected one or more remedial actions comprise at least one remedial action for modifying a configuration of the given computing resource.
12. The apparatus of claim 9 wherein the one or more assets comprise one or more health monitoring devices associated with at least one user, wherein the classified one or more patterns in the obtained time series dataset characterize one or more health conditions of the at least one user, and wherein the selected one or more remedial actions comprise at least one remedial action for alerting the at least one user to the one or more health conditions.
13. The apparatus of claim 1 wherein the one or more assets comprise two or more nodes in a communication network, wherein the classified one or more patterns in the obtained time series dataset characterize transmission of one or more signals between the two or more nodes in the communication network, and wherein the selected one or more remedial actions comprise re-transmission of the reconstructed time series dataset with the one or more imputed values to at least one additional node in the communication network.
14. The apparatus of claim 1 wherein the loss function comprises a regularized loss function comprising a first tunning parameter controlling an amount of regularization induced by variation loss and a second tuning parameter controlling an amount of regularization induced by regularization of the network parameters of the convolutional neural network, and wherein a value of the first tuning parameter is greater than a value of the second tuning parameter.
15. A computer program product comprising a non-transitory processor-readable storage medium having stored therein program code of one or more software programs, wherein the program code when executed by at least one processing device causes the at least one processing device to perform steps of:
obtaining a time series dataset having one or more missing values, the time series dataset comprising monitoring data associated with one or more assets;
generating, utilizing a machine learning algorithm, a reconstructed time series dataset having one or more imputed values for the one or more missing values in the obtained time series dataset, the machine learning algorithm comprising a generative network implementing inverse network parameter determination for network parameters of the generative network;
classifying one or more patterns in the obtained time series dataset utilizing the reconstructed time series dataset;
selecting one or more remedial actions for controlling at least one of the one or more assets based at least in part on the classified one or more patterns in the obtained time series dataset; and
initiating the selected one or more remedial actions to control the at least one asset;
wherein the machine learning algorithm comprises a convolutional neural network; and
wherein generating the reconstructed time series dataset having the one or more imputed values comprises performing two or more iterations of:
applying the obtained time series dataset having the one or more missing values to the convolutional neural network to produce a candidate reconstructed time series dataset having one or more imputed values for the one or more missing values in the obtained time series dataset; and
tuning network parameters of the convolutional neural network using a loss function that compares the candidate reconstructed time series dataset with the obtained time series dataset.
16. The computer program product of claim 15 wherein the convolutional neural network is initialized with randomized values for the network parameters.
17. The computer program product of claim 15 wherein the loss function comprises a regularized loss function comprising a first tunning parameter controlling an amount of regularization induced by a variation loss and a second tuning parameter controlling an amount of regularization induced by regularization of the network parameters of the convolutional neural network, and wherein a value of the first tuning parameter is greater than a value of the second tuning parameter.
18. A method comprising steps of:
obtaining a time series dataset having one or more missing values, the time series dataset comprising monitoring data associated with one or more assets;
generating, utilizing a machine learning algorithm, a reconstructed time series dataset having one or more imputed values for the one or more missing values in the obtained time series dataset, the machine learning algorithm comprising a generative network implementing inverse network parameter determination for network parameters of the generative network;
classifying one or more patterns in the obtained time series dataset utilizing the reconstructed time series dataset;
selecting one or more remedial actions for controlling at least one of the one or more assets based at least in part on the classified one or more patterns in the obtained time series dataset; and
initiating the selected one or more remedial actions to control the at least one asset;
wherein the machine learning algorithm comprises a convolutional neural network;
wherein generating the reconstructed time series dataset having the one or more imputed values comprises performing two or more iterations of:
applying the obtained time series dataset having the one or more missing values to the convolutional neural network to produce a candidate reconstructed time series dataset having one or more imputed values for the one or more missing values in the obtained time series dataset; and
tuning network parameters of the convolutional neural network using a loss function that compares the candidate reconstructed time series dataset with the obtained time series dataset; and
wherein the method is performed by at least one processing device comprising a processor coupled to a memory.
19. The method of claim 18 wherein the convolutional neural network is initialized with randomized values for the network parameters.
20. The method of claim 18 wherein the loss function comprises a regularized loss function comprising a first tunning parameter controlling an amount of regularization induced by a variation loss and a second tuning parameter controlling an amount of regularization induced by regularization of the network parameters of the convolutional neural network, and wherein a value of the first tuning parameter is greater than a value of the second tuning parameter.