Synthetic rare class generation by preserving morphological identity

A system (100) comprising:

• one or more data storage devices (102) operatively coupled to one or more hardware processors (104) and configured to store instructions configured for execution by the one or more hardware processors to:
• analyze a labeled abundant training dataset (+) that refers to a training dataset for a class available in more numbers and labeled rare class training dataset (−) that refers to a training dataset with limited availability compared to the training dataset;
• generate an extended synthetic rare class dataset (−−−) based on the using an extended oversampling method based on Markov chain model providing a systematic and a stochastic method to model time varying signals, wherein in the Markov chain model, a future state only depends on current state, not on predecessor states;
• and
• extract a subset of the extended synthetic rare class dataset (−−−) to obtain a synthetic rare class dataset (−−), with respect to similarity and diversity in the synthetic rare class dataset (−−) and preserving morphological identity of the synthetic rare class dataset (−−) with respect to the labeled rare class training dataset (_) while maintaining diversity and facilitates addressing class imbalance problem in applications such as identifying real-life noisy physiological time-series signals,
• wherein the subset of the extended synthetic rare class dataset is extracted without reducing dimensionality of the dataset, thereby making the step of obtaining the synthetic rare class dataset independent of dimensionality properties and is signal space rare class dataset, wherein the real-life noisy physiological time-series signals are phonocardiogram (PCG) signals.

A computer program product comprising a non-transitory computer readable medium having a computer readable program embodied therein, wherein the computer readable program, when executed on a computing device, causes the computing device to:

• analyze a labeled abundant training dataset (+) that refers to a training dataset for a class available in more numbers and labeled rare class training dataset (−) that refers to a training dataset with limited availability compared to the training dataset;
• generate an extended synthetic rare class dataset (−−−) based on the analysis using an oversampling method based on Markov chain model providing a systematic and a stochastic method to model time varying signals, wherein in the Markov chain model, a future state only depends on current state, not on predecessor states;
• and
• extract a subset of the extended synthetic rare class dataset (−−−) to obtain a synthetic rare class dataset (−−), with respect to similarity and diversity in the synthetic rare class dataset (−−) and preserving morphological identity of the synthetic rare class dataset (−−) with respect to the labeled rare class training dataset (−) while maintaining diversity and facilitates addressing class imbalance problem in applications such as identifying real-life noisy physiological time-series signals,
• wherein the subset of the extended synthetic rare class dataset is extracted without reducing dimensionality of the dataset, thereby making the step of obtaining the synthetic rare class dataset independent of dimensionality properties and is signal space rare class dataset, wherein the real-life noisy physiological time-series signals are phonocardiogram (PCG) signals.