SYSTEMS AND METHODS FOR CLASSIFICATION OF MULTI-DIMENSIONAL TIME SERIES OF PARAMETERS

20200012938

16/363038

March 25, 2019

Traditional systems and methods have implemented hand-crafted feature extraction from varying length time series that results in complexity and requires domain knowledge. Building classification models requires large labeled data and is computationally expensive. Embodiments of the present disclosure implement learning models for classification tasks in multi-dimensional time series by performing feature extraction from entity's parameters via unsupervised encoder and build a non-temporal linear classifier model. A fixed-dimensional feature vector is outputted using a pre-trained unsupervised encoder, which acts as off-the shelf feature extractor. Extracted features are concatenated to learn a non-temporal linear classification model and weight is assigned to each extracted feature during learning which helps to determine relevant parameters for each class. Mapping from parameters to target class is considered while constraining the linear model to use only subset of large number of features.

Tata Consultancy Services Limited (Mumbai, IN)

1. A processor implemented method, comprising: obtaining, via one or more hardware processors, a unique time series data corresponding to a plurality of parameters of one or more entities (202); automatically extracting, using an unsupervised encoder integrated within a Deep Recurrent Neural Network (RNN) executed by the one or more hardware processors, one or more features from the unique time series to obtain a unique features set for each of the plurality of parameters, wherein the unique features set comprises a fixed-dimensional feature vector (204); concatenating the one or more extracted features from the unique features set pertaining each of the plurality of parameters to obtain a concatenated features set comprising a fixed-dimensional concatenated feature vector (206); learning a non-temporal linear classification model based on the concatenated features set, wherein during the learning of the non-temporal linear classification model a weight is assigned to each feature from the concatenated features set (208); and generating a relevance score for each of the plurality of parameters based on the weight of each feature from the concatenated features set to validate the learned non-temporal linear classification model (210).

2. The processor implemented method of claim 1, further comprising: receiving an input time series corresponding to the plurality of parameters of the entities (212); automatically extracting one or more features from the input time series (214); and applying the validated learned classification model on the input time series based on the extracted one or more features to obtain a class for the input time series corresponding to the plurality of parameters of the entities (216).

3. The processor implemented method of claim 2, wherein the input time series and the unique time series data are a fixed length data or a variable length data.

4. The processor implemented method of claim 1, wherein the weight is obtained using a Least Absolute Shrinkage and Selection Operator (LASSO)-regularized loss function.

5. A system comprising: a memory (102) storing instructions; one or more communication interfaces (106); and one or more hardware processors (104) coupled to the memory (102) via the one or more communication interfaces (106), wherein the one or more hardware processors (104) are configured by the instructions to: obtain a unique time series data corresponding to a plurality of parameters of one or more entities; automatically extract, using an unsupervised encoder integrated within a Deep Recurrent Neural Network (RNN) implemented by the system 100, one or more features from the unique time series to obtain a unique features set for each of the plurality of parameters, wherein the unique features set comprises a fixed-dimensional feature vector; concatenate the one or more extracted features from the unique features set pertaining each of the plurality of parameters to obtain a concatenated features set comprising a fixed-dimensional concatenated feature vector; learn a non-temporal linear classification model based on the concatenated features set, wherein during the learning of the non-temporal linear classification model a weight is assigned to each feature from the concatenated features set; and generate a relevance score for each of the plurality of parameters based on the weight of each feature from the concatenated features set to validate the learned non-temporal linear classification model.

6. The system of claim 5, wherein the one or more hardware processors are further configured to: receive an input time series corresponding to the plurality of parameters of the entities; automatically extract one or more features from the input time series; apply the validated learned classification model on the input time series based on the extracted one or more features to obtain a class for the input time series corresponding to the plurality of parameters of the entities.

7. The system of claim 5, wherein the input time series and the unique time series data are a fixed length data or a variable length data.

8. The system of claim 5, wherein the weight is obtained using a Least Absolute Shrinkage and Selection Operator (LASSO)-regularized loss function.

9. One or more non-transitory machine readable information storage mediums comprising one or more instructions which when executed by one or more hardware processors cause classifying multi-dimensional time series of parameters by: obtaining, via one or more hardware processors, a unique time series data corresponding to a plurality of parameters of one or more entities; automatically extracting, using an unsupervised encoder integrated within a Deep Recurrent Neural Network (RNN) executed by the one or more hardware processors, one or more features from the unique time series to obtain a unique features set for each of the plurality of parameters, wherein the unique features set comprises a fixed-dimensional feature vector; concatenating the one or more extracted features from the unique features set pertaining each of the plurality of parameters to obtain a concatenated features set comprising a fixed-dimensional concatenated feature vector; learning a non-temporal linear classification model based on the concatenated features set, wherein during the learning of the non-temporal linear classification model a weight is assigned to each feature from the concatenated features set; and generating a relevance score for each of the plurality of parameters based on the weight of each feature from the concatenated features set to validate the learned non-temporal linear classification model.

10. The one or more non-transitory machine readable information storage mediums of claim 9, wherein the instructions which when executed by the one or more hardware processors further cause: receiving an input time series corresponding to the plurality of parameters of the entities; automatically extracting one or more features from the input time series; and applying the validated learned classification model on the input time series based on the extracted one or more features to obtain a class for the input time series corresponding to the plurality of parameters of the entities.

11. The one or more non-transitory machine readable information storage mediums of claim 10, wherein the input time series and the unique time series data are a fixed length data or a variable length data.

12. The one or more non-transitory machine readable information storage mediums of claim 9, wherein the weight is obtained using a Least Absolute Shrinkage and Selection Operator (LASSO)-regularized loss function.

06; 06

edspap.20200012938