US 12,443,875 B2
Explanatory confusion matrices for machine learning
Alex Swain, Cedar Park, TX (US); Stefan A. G. Van Der Stockt, Austin, TX (US); Edward James Biddle, Winchester (GB); and Daniel Kuehn, Austin, TX (US)
Assigned to INTERNATIONAL BUSINESS MACHINES CORPORATION, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on Nov. 10, 2020, as Appl. No. 17/093,978.
Prior Publication US 2022/0147862 A1, May 12, 2022
Int. Cl. G06N 20/20 (2019.01); G06F 18/40 (2023.01); G06N 5/01 (2023.01); G06N 5/04 (2023.01); G06N 20/00 (2019.01)
CPC G06N 20/00 (2019.01) 9 Claims
OG exemplary drawing
 
1. A computer implemented method for creating explanatory confusion matrices, comprising:
building, by one or more computer processors with access to classification results from a machine learning model, a user interface including a confusion matrix having a plurality of cells for visualizing the classification results, wherein the plurality of cells includes respective electronic links to a corresponding set of data that produced the classification results visualized in the plurality of cells, and wherein each cell of the plurality of cells that is associated with a respective incorrect classification by the machine learning model is selectable to execute an explanatory application that is configured to identify at least one data-specific reason that each misclassified data resulting in the respective incorrect classification is incorrectly classified and at least one recommendation to correct the machine learning model;
executing, based on a user manipulation of the user interface to select a specific cell in the confusion matrix that is associated with the respective incorrect classification by the machine learning model, the explanatory application to identify the at least one data-specific reason that each misclassified data resulting in the respective incorrect classification is incorrectly classified and the at least one recommendation to correct the machine learning model;
visualizing, by the one or more computer processors on the user interface, the at least one data-specific reason that each misclassified data associated with the specific cell is incorrectly classified and the at least one recommendation to correct the machine learning model;
causing, based on the user manipulation of the user interface to correct the respective incorrect classification associated with the specific cell based on the at least one recommendation displayed on the user interface, at least one correction to each misclassified data associated with the specific cell; and
responsive to the user manipulation of the user interface to correct each misclassified data associated with the specific cell, generating, by the one or more computer processors, correctly classified data and retraining, by the one or more computer processors, the machine learning model with the correctly classified data to adjust at least one feature of the machine learning model such that the machine learning model retrained to include the at least one adjusted feature correctly classifies each misclassified data associated with the specific cell and improves future predictions on unseen data.