Prediction enhancement for classification models
Systems and methods provide reception of an identifier of a machine learning classification model and a prediction generated by the machine learning classification model, identification of model configuration data associated with the machine learning classification model, modification of the prediction based on the model configuration data to generate an enhanced prediction comprising calibrated probabilities, and returning of the enhanced prediction.
1 . A system comprising:
a storage device; and
a processor to execute processor-executable program code stored on the storage device to cause the system to:
receive an identifier of a machine learning classification model and a prediction generated by the machine learning classification model;
identify model configuration data associated with the machine learning classification model;
modify the prediction based on the model configuration data to generate an enhanced prediction comprising calibrated probabilities;
modify the enhanced prediction based on an indication of similarities between classes of the machine learning classification model included in the model configuration data, the modification of the enhanced prediction comprising a determination, based on the similarities, of a first class of the classes associated with a highest probability and a second class associated with a highest probability of all classes which are not similar to the first class, and normalization of the probabilities associated with the second class and all classes associated with a lower probability than the second class and
return the enhanced prediction.
2 . A system according to claim 1 , the processor to execute processor- executable program code stored on the storage device to cause the system to:
modify the enhanced prediction based on the model configuration data to adapt the enhanced prediction to a balanced input data set.
3 . A system according to claim 2 ,
wherein the model configuration data comprises a value associated with each class of the machine learning classification model which represents a number of observations corresponding to that class in a training set used to train the machine learning classification model, and
wherein modification of the enhanced prediction is based on the values associated with each class of the machine learning classification model.
4 . A system according to claim 2 , the processor to execute processor-executable program code stored on the storage device to cause the system to:
modify the modified enhanced prediction based on similarities between classes of the machine learning classification model,
wherein the model configuration data comprises the similarities.
5 . A system according to claim 4 , wherein modification of the modified enhanced prediction comprises determination, based on the similarities, of a first class of the classes associated with a highest probability and a second class associated with a highest probability of all classes which are not similar to the first class, and normalization of the probabilities associated with the second class and all classes associated with a lower probability than the second class.
6 . A computer-implemented method comprising:
receiving an identifier of a machine learning classification model and a prediction generated by the machine learning classification model;
identifying model configuration data associated with the machine learning classification model;
modifying the prediction based on the model configuration data to generate an enhanced prediction comprising calibrated probabilities;
modifying the enhanced prediction based on an indication of similarities between classes of the machine learning classification model included in the model configuration data, the modifying of the enhanced prediction including determining, based on the similarities, of a first class of the classes associated with a highest probability and a second class associated with a highest probability of all classes which are not similar to the first class and normalizing the probabilities associated with the second class and all classes associated with a lower probability than the second class; and
returning the enhanced prediction.
7 . A method according to claim 6 , further comprising:
modifying the enhanced prediction based on the model configuration data to adapt the enhanced prediction to a balanced input data set.
8 . A method according to claim 7 ,
wherein the model configuration data comprises a value associated with each class of the machine learning classification model which represents a number of observations corresponding to that class in a training set used to train the machine learning classification model, and
wherein modification of the enhanced prediction is based on the values associated with each class of the machine learning classification model.
9 . A method according to claim 7 , further comprising:
modifying the modified enhanced prediction based on similarities between classes of the machine learning classification model,
wherein the model configuration data comprises the similarities.
10 . A method according to claim 9 , wherein modifying the modified enhanced prediction comprises:
determining, based on the similarities, of a first class of the classes associated with a highest probability and a second class associated with a highest probability of all classes which are not similar to the first class, and
normalizing the probabilities associated with the second class and all classes associated with a lower probability than the second class.
11 . A non-transitory medium storing processor-executable program code, the program code executable to cause a system to:
receive an identifier of a machine learning classification model and a prediction generated by the machine learning classification model;
identify model configuration data associated with the identifier of the machine learning classification model, the model configuration data comprising a calibration function;
modify the prediction based on the calibration function to generate an enhanced prediction comprising calibrated probabilities;
modify the modified enhanced prediction based on an indication of similarities between classes of the machine learning classification model included in the model configuration data, the modifying of the modified enhanced prediction comprising a determination, based on the similarities, of a first class of the classes associated with a highest probability and a second class associated with a highest probability of all classes which are not similar to the first class, and normalization of the probabilities associated with the second class and all classes associated with a lower probability than the second class; and
return the enhanced prediction.
12 . A medium according to claim 11 , the program code executable to cause a system to:
modify the enhanced prediction based on the model configuration data to adapt the enhanced prediction to a balanced input data set.
13 . A medium according to claim 12 ,
wherein the model configuration data comprises a value associated with each class of the machine learning classification model which represents a number of observations corresponding to that class in a training set used to train the machine learning classification model, and
wherein modification of the enhanced prediction is based on the values associated with each class of the machine learning classification model.
14 . A medium according to claim 11 , the program code executable to cause a system to:
modify the enhanced prediction based on similarities between classes of the machine learning classification model,
wherein the model configuration data comprises the similarities.