Method for automatic estimation of spatio-temporal entity counts using machine learning from partially observable location data

A system for automatically estimating spatio-temporal entity counts in real time and for a future time window using machine learning from partially observable location data, the system comprises:

• a data aggregator that is configured to obtain in real time a plurality of data streams with a unique entity identifier from independently controlled sources, wherein the plurality of data streams comprises timestamp data and location indexed data that partially characterizes an activity of an entity associated with the unique entity identifier;
• a geocoder that is configured to transform the plurality of data streams into geolocation of the entity;
• a geolocation mapper that is configured to identify corresponding geolocation of the entity in a hyper-cube computational data structure that is spatially partitioned with geo hashes in terms of a key value data structure for a plurality of geolocations by mapping the geolocation of the entity with the key value data structure in the hyper-cube computational data structure;
• a key value data structure updater that is configured to update in real time the key value data structure that corresponds to geolocation of the entity, with the unique entity identifier and the timestamp data;
• a hyper cube estimator that is configured to determine a lower bound number of an entity count for each or combinations of a plurality of spatio temporal dimensions based on the unique entity identifier and the timestamp data updated in the geolocation of the key value data structure;
• a census-based extrapolator that is configured to determine an upper bound number of the entity count for each or combinations of the plurality of spatio temporal dimensions by extrapolating recent census data;
• an entity estimator that is configured to (i) determine an entity count for each or combinations of the plurality of spatio temporal dimensions in real time by combining the lower bound number and the upper bound number of the entity count using a machine learning based time series model, and (ii) estimate, using the machine learning based time series model, spatio temporal entity count for a future time window in response to a query criterion.

A one or more non-transitory computer-readable storage mediums storing the one or more sequences of instructions, which when executed by the one or more processors, causes to perform a method of automatically estimating spatio-temporal entity counts in real time and for a future time window using machine learning from partially observable location data, wherein the method comprises:

• obtaining, in real time, a plurality of data streams with a unique entity identifier from independently controlled sources, wherein the plurality of data streams comprises timestamp data and location indexed data that partially characterizes an activity of an entity associated with the unique entity identifier;
• transforming, using a geocoder, the plurality of data streams into geolocation of the entity;
• identifying, using a geolocation mapper, corresponding geolocation of the entity in a hyper-cube computational data structure that is spatially partitioned with geo hashes in terms of a key value data structure for a plurality of geolocations by mapping the geolocation of the entity with the key value data structure in the hyper-cube computational data structure;
• updating, in real time, the key value data structure that corresponds to geolocation of the entity, with the unique entity identifier and the timestamp data;
• determining, using a hyper cube estimator, a lower bound number of an entity count for each or combinations of a plurality of spatio temporal dimensions based on the unique entity identifier and the timestamp data updated in the geolocation of the key value data structure;
• determining, using a census-based extrapolator, an upper bound number of the entity count for each or combinations of the plurality of spatio temporal dimensions by extrapolating recent census data;
• combining, using an entity estimator, the lower bound number and the upper bound number of the entity count for determining an entity count for each or combinations of the plurality of spatio temporal dimensions in real time using a machine learning based time series model;
• and
• estimating, using the machine learning based time series model, spatio temporal entity count for a future time window in response to a query criterion.