Interpretation of machine learning classifications in clinical diagnostics using shapely values and uses thereof

A method for generating an explainable machine learning model in clinical diagnostics, comprising the steps of:

• a) conducting a physical measurement process on a multitude of biological samples obtained from humans, the samples and associated clinical data forming a classifier development set, thereby obtaining a set of values for a multitude of attributes for each member of the development set,
• b) storing the set of values in computer memory;
• c) with a computer, training a classifier arranged as a logistical combination of atomic classifiers with drop-out regularization from set of feature values for the development set;
• and
• d) using a Monte-Carlo sampling method to sample the classification of subsets of the attributes to calculate Shapley values indicating the relative importance of some or all of the attributes to classifications of the development set produced by the classifier trained in step c) without retraining the classifier.

A method for generating an explainable machine learning model in clinical diagnostics, comprising the steps of:

• a) conducting a physical measurement process on a multitude of biological samples obtained from humans, the samples and associated clinical data forming a classifier development set, thereby obtaining a set of values for a multitude of attributes for each member of the development set;
• b) storing the set of feature values in computer memory;
• c) with a computer, training a classifier arranged as a logistical combination of atomic classifiers with drop-out regularization from set of values for the development set;
• and
• d) calculating Shapley values exactly indicating the relative importance of some or all of the attributes to classifications of the development set produced by the classifier trained in step c) without retraining the classifier, wherein the exact calculation is performed using equations 5.2 and 5.3 and 5.4 or the equivalent.