Event streams in a machine learning system for demand forecasting

A system comprising:

• a computer hardware server comprising a processor, system memory and a network interface for communicating via one or more computer networks, the computer hardware server implementing a machine learning system having access to one or more databases adapted for storing data routed over the one or more computer networks from one or more data sources, the machine learning system configured to perform operations for predicting the impact of a plurality of events on a demand forecast for an organization using event streams, the operations comprising:
• receiving, via the computer networks, demand data for products or services of the organization from a database of demand data and events data comprising event attributes from a database of events data for a plurality of different events located within a distance from the geolocation of the organization;
• performing a feature selection process for selecting, from among a set of features that are available for use in training or updating a final machine learning model to compute one or more forecasts for demand of products or services provided by an organization, a subset of the features to use for training the final machine learning model, comprising:
• receiving geolocation data identifying the geolocation of the organization;
• generating a range of distances based on segmenting the distance from the geolocation of the organization into a plurality of different distances;
• normalizing event categories for each of the plurality of different events by assigning them to one of a set of predefined event categories;
• grouping the events into different combinations of the events based on event attributes, wherein events in a same predefined event category and within a same range of distances from the geolocation of the organization are grouped together;
• and
• generating a set of candidate event streams from the plurality of different events based on executing a feature selection process comprising:
• (i) computing, for each of the set of candidate event streams, an impact that the candidate event stream has on an initial set of demand forecast values using a trial machine learning model that is trained to compute, for an input candidate event stream, the impact that the input candidate event stream has on the initial set of demand forecast values;
• (ii) evaluating a magnitude of the impacts of each of the candidate event streams on the computed initial set of demand forecast values that were computed using the trial machine learning model to determine which candidate event streams have an impact on the computed demand forecast values;
• and
• (iii) filtering the candidate event streams to obtain a filtered set of event streams comprising ones of the candidate event streams determined to have an impact on the computed initial set of demand forecast values;
• and
• using the filtered set of event streams and the demand data as training data to train or update the final machine learning model to compute the one or more forecasts for demand of products or services provided by the organization.

A non-transitory computer readable storage medium adapted for storing programmed computer code executable by a computer hardware server for implementing a machine learning system for performing operations for predicting the impact of a plurality of events on a demand forecast for an organization using event streams, the operations comprising:

• receiving, via the computer networks, demand data for products or services of the organization from a database of demand data and events data comprising event attributes from a database of events data for a plurality of different events located within a distance from the geolocation of the organization;
• performing a feature selection process for selecting, from among a set of features that are available for use in training or updating a final machine learning model to compute one or more forecasts for demand of products or services provided by an organization, a subset of the features to use for training the final machine learning model, comprising:
• receiving geolocation data identifying the geolocation of the organization;
• generating a range of distances based on segmenting the distance from the geolocation of the organization into a plurality of different distances;
• normalizing event categories for each of the plurality of different events by assigning them to one of a set of predefined event categories;
• grouping the events into different combinations of the events based on event attributes, wherein events in a same predefined event category and within a same range of distances from the geolocation of the organization are grouped together;
• generating a set of candidate event streams from the plurality of different events based on executing a feature selection process comprising:
• (i) computing, for each of the set of candidate event streams, an impact that the candidate event stream has on an initial set of demand forecast values using a trial machine learning model that is trained to compute, for an input candidate event stream, the impact that the input candidate event stream has on the initial set of demand forecast values;
• (ii) evaluating a magnitude of the impacts of each of the candidate event streams on the computed initial set of demand forecast values that were computed using the trial machine learning model to determine which candidate event streams have an impact on the computed demand forecast values;
• and
• (iii) filtering the candidate event streams to obtain a filtered set of event streams comprising ones of the candidate event streams determined to have an impact on the computed initial set of demand forecast values;
• and
• using the filtered set of event streams and the demand data as training data to train or update the final machine learning model to compute the one or more forecasts for demand of products or services provided by the organization.