Mutual information with absolute dependency for feature selection in machine learning models

A system comprising:

• one or more computers;
• and
• one or more computer-readable storage devices storing instructions that, when executed by the one or more computers, cause the one or more computers to perform operations comprising:
• receiving, by the one or more computers, a dataset comprising multiple items of a same type, wherein each item has multiple properties;
• determining, for each of multiple iterations, a value representing mutual information between a first property and a second property, wherein, for each iteration, determining the value comprises:
• selecting, by the one or more computers, one of the multiple properties as the first property and another of the multiple properties as the second property;
• and
• determining, by the one or more computers, the value representing the mutual information between the first property and the second property that are selected for the iteration, wherein the value representing the mutual information is generated using a summation of terms that are generated using logarithm values, and wherein the values of the terms are generated using an absolute value operation that disallows negative logarithm values from decreasing the summation;
• based on the determined values representing mutual information between the selected properties from the multiple properties, selecting, by the one or more computers of the system, a subset of the multiple properties for use with a machine learning model;
• receiving, by the one or more computers, another item of the same type as the multiple items from the dataset, wherein the other item has a known value for at least one of the subset of properties;
• predicting, by the one or more computers, a value for a property of the other item having an unknown value, wherein the value is predicted using the machine learning model based on the known value for the at least one of the subset of properties;
• and
• providing, by the one or more computers, the predicted value over a network to a client device for display on a user interface of the client device.

One or more non-transitory computer-readable media storing instructions that, when executed by one or more computers of a machine learning system, cause the one or more computers to perform operations comprising:

• receiving, by the one or more computers of the machine learning system, a dataset comprising multiple items of a same type, wherein each item has multiple properties;
• determining, for each of multiple iterations, a value representing mutual information between a first property and a second property, wherein, for each iteration, determining the value comprises:
• selecting, by the one or more computers of the machine learning system, one of the multiple properties as the first property and another of the multiple properties as the second property;
• and
• determining, by the one or more computers of the machine learning system, the value representing the mutual information between the first property and the second property that are selected for the iteration, wherein the value representing the mutual information is generated using a summation of terms that are generated using logarithm values, and wherein the values of the terms are generated using an absolute value operation that disallows negative logarithm values from decreasing the summation;
• based on the determined values representing mutual information between the selected properties from the multiple properties, selecting, by the one or more computers of the machine learning system, a subset of the multiple properties for use in the machine learning system;
• receiving, by the one or more computers of the machine learning system, another item of the same type as the multiple items from the dataset, wherein the other item has a known value for at least one of the subset of properties;
• predicting, by the one or more computers of the machine learning system, a value for a property of the other item having an unknown value, wherein the value is predicted using a machine learning model of the machine learning system based on the known value for the at least one of the subset of properties;
• and
• providing, by the one or more computers of the machine learning system, the predicted value over a network to a client device for display on a user interface of the client device.