Onboarding of return path data providers for audience measurement

1. An audience measurement computing system to predict return path data quality, the audience measurement computing system comprising:

media metering devices installed at panelist households statistically-selected and recruited by an audience measurement entity, the media metering devices programmed by the audience measurement entity to monitor media played in the panelist households and automatically report meter tuning data to a network-connected server of the audience measurement computing system via respective network interfaces of the media metering devices, wherein the panelist households are return path data subscribers, wherein the media metering devices monitor media played from first return path devices and first non-return path devices at the panelist households and capture, as part of the meter tuning data, a total number of minutes for the media that was monitored for the panelist households, wherein the first return path devices include first set-top boxes (STBs) connected to or integrated with televisions of the panelist households, wherein non-panelist households that are return path data subscribers include second return path devices and second non-return path devices, and wherein the second return path devices include second STBs connected to or integrated with televisions of the non-panelist households;

the network-connected server, comprising:

a network interface;

at least one memory storing computer readable instructions; and

at least one processor to execute the computer readable instructions to perform operations comprising:

automatically obtaining, via the network interface, the meter tuning data from the media metering devices;

obtaining, via the network interface, from the first STBs, first return path tuning data included in first return path data, wherein the first STBs are configured to collect the first return path data and report the first return path data to the network-connected server, the first return path data comprising tuning events and commands detected by the first STBs, the tuning events comprising a channel change and a start of a media presentation, and the commands comprising a power on command and a power off command, wherein the first return path devices capture, as part of the first return path tuning data, a total number of return path device tuning data minutes for the panelist households;

obtaining, via the network interface, from the second STBs, second return path tuning data included in second return path data, wherein the second STBs are configured to collect the second return path data and report the second return path data to the network-connected server; 'comparing the minutes of the meter tuning data captured by the media metering devices and the minutes of the first return path tuning data captured by the first return path devices of the panelist households to determine missing data rates indicative of a quantity of the meter tuning data from the panelist households that is not included in the corresponding first return path tuning data of the panelist households;

computing a first data set of model features from validation tuning data reported from the media metering devices, the validation tuning data comprising return path tuning data from the first return path tuning data for which the network-connected server determined a missing data rate, the first data set of first model features being predictive of return path data that is missing from the non-panelist households;

computing a second data set of model features from the second return path tuning data;

based on the first data set, training a machine learning model to, for each of the non-panelist households and relative to a particular period of time, (i) predict a quality indicator of a particular return path data reported from the second STBs of that non-panelist household for the particular period of time and (ii) determine, based on the predicted quality indicator, whether to remove the particular return path tuning data from further processing for the particular period of time, wherein the machine learning model comprises a neural network;

applying the trained machine learning model to the second data set to, for each of the non-panelist households and relative to the particular period of time, (i) predict the quality indicator of the second return path data for the particular period of time and (ii) determine, based on the predicted quality indicator, whether to remove the second return path tuning data from further processing for the particular period of time, wherein determining, based on the predicted quality indicator, whether to remove the second return path tuning data from further processing for the particular period of time comprises determining that the second return path tuning data should be removed from further processing for the particular period of time; and

based on the determination that the second return path tuning data should be removed from further processing for the particular period of time, generating ratings for a media presentation without processing the second return path tuning data, the ratings corresponding to the particular period of time.

2. The audience measurement computing system of claim 1 , the operations further comprising:

before applying the trained machine learning model, splitting a training data set using cross validation with numTrees and maxDepth parameters, the numTrees parameter representing a number of decision trees in a random forest classifier of the machine learning model, and the maxDepth parameter representing a maximum number of levels in each decision tree in the random forest classifier.

3. The audience measurement computing system of claim 1 , wherein training the machine learning model comprises training a random forest machine learning model with a k-fold cross validation.

4. The audience measurement computing system of claim 1 , wherein the operations are performed daily, and

wherein the particular period of time is a particular day.

5. The audience measurement computing system of claim 1 , the operations further comprising:

determining an onboarding status for a return path data provider that is to provide media to be accessed by the second STBs, the onboarding status based on the predicted quality indicator of the second return path data.

6. The audience measurement computing system of claim 1 , the operations further comprising:

filtering a portion of the meter tuning data, a portion of the first return path tuning data, and a portion of the second return path tuning data that are not associated with a first viewing period.

7. A non-transitory computer readable medium comprising instructions that, when executed, cause a processor to perform operations comprising:

automatically obtaining, via a network interface of a network-connected server, meter tuning data from media metering devices, the media metering devices installed at panelist households statistically-selected and recruited by an audience measurement entity, the media metering devices programmed by the audience measurement entity to monitor media played in the panelist households and automatically report the meter tuning data to the network-connected server of the audience measurement computing system via respective network interfaces of the media metering devices, wherein the panelist households are return path data subscribers, wherein the media metering devices monitor media played from first return path devices and first non-return path devices at the panelist households and capture, as part of the meter tuning data, a total number of minutes for the media that was monitored for the panelist households, wherein the first return path devices include first set-top boxes (STBs) connected to or integrated with televisions of the panelist households, wherein non-panelist households that are return path data subscribers include second return path devices and second non-return path devices, and wherein the second return path devices include second STBs connected to or integrated with televisions of the non-panelist households;

obtaining, via the network interface, from the first STBs, first return path tuning data included in first return path data, wherein the first STBs are configured to collect the first return path data and report the first return path data to the network-connected server, the first return path data comprising tuning events and commands detected by the first STBs, the tuning events comprising a channel change and a start of a media presentation, and the commands comprising a power on command and a power off command, wherein the first return path devices capture, as part of the first return path tuning data, a total number of return path device tuning data minutes for the panelist households;

obtaining, via the network interface, from the second STBs, second return path tuning data included in second return path data, wherein the second STBs are configured to collect the second return path data and report the second return path data to the network-connected server;

comparing the minutes of the meter tuning data captured by the media metering devices and the minutes of the first return path tuning data captured by the first return path devices of the panelist households to determine missing data rates indicative of a quantity of the meter tuning data from the panelist households that is not included in the corresponding first return path tuning data of the panelist households;

computing a first data set of model features from validation tuning data reported from the media metering devices, the validation tuning data comprising return path tuning data from the first return path tuning data for which the network-connected server determined a missing data rate, the first data set of first model features being predictive of return path data that is missing from the non-panelist households;

computing a second data set of model features from the second return path tuning data;

based on the first data set, training a machine learning model to, for each of the non-panelist households and relative to a particular period of time, (i) predict a quality indicator of a particular return path data reported from the second STBs of that non- panelist household for the particular period of time and (ii) determine, based on the predicted quality indicator, whether to remove the particular return path tuning data from further processing for the particular period of time, wherein the machine learning model comprises a neural network;

applying the trained machine learning model to the second data set to, for each of the non-panelist households and relative to the particular period of time, (i) predict the quality indicator of the second return path data for the particular period of time and (ii) determine, based on the predicted quality indicator, whether to remove the second return path tuning data from further processing for the particular period of time, wherein determining, based on the predicted quality indicator, whether to remove the second return path tuning data from further processing for the particular period of time comprises determining that the second return path tuning data should be removed from further processing for the particular period of time; and

based on the determination that the second return path tuning data should be removed from further processing for the particular period of time, generating ratings for a media presentation without processing the second return path tuning data, the ratings corresponding to the particular period of time.

8. The non-transitory computer readable medium of claim 7 , the operations further comprising:

before applying the trained machine learning model, splitting a training data set using cross validation with numTrees and maxDepth parameters, the numTrees parameter representing a number of decision trees in a random forest classifier of the machine learning model, and the maxDepth parameter representing a maximum number of levels in each decision tree in the random forest classifier.

9. The non-transitory computer readable medium of claim 7 , wherein training the machine learning model comprises training a random forest machine learning model with a k-fold cross validation.

10. The non-transitory computer readable medium of claim 7 , wherein the operations are performed daily, and

wherein the particular period of time is a particular day.

11. The non-transitory computer readable medium of claim 7 , the operations further comprising:

determining an onboarding status for a return path data provider that is to provide media to be accessed by the second STBs, the onboarding status based on the predicted quality indicator of the second return path data.

12. The non-transitory computer readable medium of claim 7 , the operations further comprising:

filtering a portion of the meter tuning data, a portion of the first return path tuning data, and a portion of the second return path tuning data that are not associated with a first viewing period.

13. A method comprising:

Automatically obtaining, via a network interface of a network-connected server, meter tuning data from media metering devices, the media metering devices installed at panelist households statistically-selected and recruited by an audience measurement entity, the media metering devices programmed by the audience measurement entity to monitor media played in the panelist households and automatically report the meter tuning data to the network-connected server of the audience measurement computing system via respective network interfaces of the media metering devices, wherein the panelist households are return path data subscribers, wherein the media metering devices monitor media played from first return path devices and first non-return path devices at the panelist households and capture, as part of the meter tuning data, a total number of minutes for the media that was monitored for the panelist households, wherein the first return path devices include first set-top boxes (STBs) connected to or integrated with televisions of the panelist households, wherein non-panelist households that are return path data subscribers include second return path devices and second non-return path devices, and wherein the second return path devices include second STBs connected to or integrated with televisions of the non-panelist households;

obtaining, via the network interface, from the first STBs, first return path tuning data included in first return path data, wherein the first STBs are configured to collect the first return path data and report the first return path data to the network-connected server, the first return path data comprising tuning events and commands detected by the first STBs, the tuning events comprising a channel change and a start of a media presentation, and the commands comprising a power on command and a power off command, wherein the first return path devices capture, as part of the first return path tuning data, a total number of return path device tuning data minutes for the panelist households;

obtaining, via the network interface, from the second STBs, second return path tuning data included in second return path data, wherein the second STBs are configured to collect the second return path data and report the second return path data to the network-connected server;

comparing the minutes of the meter tuning data captured by the media metering devices and the minutes of the first return path tuning data captured by the first return path devices of the panelist households to determine missing data rates indicative of a quantity of the meter tuning data from the panelist households that is not included in the corresponding first return path tuning data of the panelist households;

computing a first data set of model features from validation tuning data reported from the media metering devices, the validation tuning data comprising return path tuning data from the first return path tuning data for which the network-connected server determined a missing data rate, the first data set of first model features being predictive of return path data that is missing from the non-panelist households;

computing a second data set of model features from the second return path tuning data;

based on the first data set, training a machine learning model to, for each of the non-panelist households and relative to a particular period of time, (i) predict a quality indicator of a particular return path data reported from the second STBs of that non-panelist household for the particular period of time and (ii) determine, based on the predicted quality indicator, whether to remove the particular return path tuning data from further processing for the particular period of time, wherein the machine learning model comprises a neural network;

applying the trained machine learning model to the second data set to, for each of the non-panelist households and relative to the particular period of time, (i) predict the quality indicator of the second return path data for the particular period of time and (ii) determine, based on the predicted quality indicator, whether to remove the second return path tuning data from further processing for the particular period of time, wherein determining, based on the predicted quality indicator, whether to remove the second return path tuning data from further processing for the particular period of time comprises determining that the second return path tuning data should be removed from further processing for the particular period of time; and

based on the determination that the second return path tuning data should be removed from further processing for the particular period of time, generating ratings for a media presentation without processing the second return path tuning data, the ratings corresponding to the particular period of time.

14. The method of claim 13 , further comprising:

before applying the trained machine learning model, splitting a training data set using cross validation with numTrees and maxDepth parameters, the numTrees parameter representing a number of decision trees in a random forest classifier of the machine learning model, and the maxDepth parameter representing a maximum number of levels in each decision tree in the random forest classifier.

15. The method of claim 13 , wherein training the machine learning model comprises training a random forest machine learning model with a k-fold cross validation.

16. The method of claim 13 , wherein the method is performed daily, and wherein the particular period of time is a particular day.

17. The method of claim 13 , further comprising:

determining an onboarding status for a return path data provider that is to provide media to be accessed by the second STBs, the onboarding status based on the predicted quality indicator of the second return path data.

18. The method of claim 13 , further comprising:

filtering a portion of the meter tuning data, a portion of the first return path tuning data, and a portion of the second return path tuning data that are not associated with a first viewing period.