Classifier processing using multiple binary classifier stages

A computer program product, the computer program product comprising one or more computer readable storage media, and program instructions collectively stored on the one or more computer readable storage media, the program instructions executable by one or more processors to cause the one or more processors to perform operations comprising:

• mitigating a classification bias induced by an imbalanced representation of a plurality of classes in training data, the mitigation comprising:
• generating a training batch of data points from the training data;
• building a multi-class classifier having a series of binary classifiers arranged in a first order, wherein each of the series of binary classifiers is associated with a respective class of the plurality of classes;
• training the multi-class classifier with the series of binary classifiers being arranged in the first order using the training batch of data points, the training comprising, at each binary classifier in succession;
• identifying a data point as belonging to a class from the plurality of classes resulting in a classified data point;
• updating, prior to the training batch being inputted into a subsequent binary classifier, the training batch to exclude the classified data point from the training batch to form an updated training batch;
• inputting the updated training batch to the subsequent binary classifier, the updating and inputting steps preventing a reclassification of the classified data point by the subsequent binary classifier thereby mitigating the classification bias, wherein the reclassification is a source of the classification bias intensification;
• modifying the multi-class classifier by changing the order of the series of binary classifiers resulting in the series of binary classifiers being arranged in a second order;
• repeating the training, the identifying, the updating, and the inputting steps using the multi-class classifier with the series of binary classifiers being arranged in the second order using the training batch of data points;
• and
• selecting a final configuration of the multi-class classifier based at least in part on a comparison of first training results from training with the series of binary classifiers arranged in the first order to second training results from training with the series of binary classifiers arranged in the second order.

A computer system comprising one or more processors and one or more computer readable storage media, and program instructions collectively stored on the one or more computer readable storage media, the program instructions executable by the one or more processors to cause the one or more processors to perform operations comprising:

• mitigating a classification bias induced by an imbalanced representation of a plurality of classes in training data, the mitigation comprising:
• generating a training batch of data points from the training data;
• building a multi-class classifier having a series of binary classifiers arranged in a first order, wherein each of the series of binary classifiers is associated with a respective class of the plurality of classes;
• training the multi-class classifier with the series of binary classifiers being arranged in the first order using the training batch of data points, the training comprising, at each binary classifier in succession;
• identifying a data point as belonging to a class from the plurality of classes resulting in a classified data point;
• updating, prior to the training batch being inputted into a subsequent binary classifier, the training batch to exclude the classified data point from the training batch to form an updated training batch;
• inputting the updated training batch to the subsequent binary classifier, the updating and inputting steps preventing a reclassification of the classified data point by the subsequent binary classifier thereby mitigating the classification bias, wherein the reclassification is a source of the classification bias intensification;
• modifying the multi-class classifier by changing the order of the series of binary classifiers resulting in the series of binary classifiers being arranged in a second order;
• repeating the training, the identifying, the updating, and the inputting steps using the multi-class classifier with the series of binary classifiers being arranged in the second order using the training batch of data points;
• and
• selecting a final configuration of the multi-class classifier based at least in part on a comparison of first training results from training with the series of binary classifiers arranged in the first order to second training results from training with the series of binary classifiers arranged in the second order.