Machine learning technologies for predicting dwell time
Systems, methods, and computer-readable medium storing instructions for using a machine learning to predict dwell times for shipments along routes are described herein. The method, system, or instructions may include training the machine learning model using training data, storing the machine learning model in a memory, accessing route data of a shipment of interest, analyzing the route data using the machine learning model, and, based on the analyzing of the route data, outputting a predicted dwell time at a stop of the shipment of interest.
1 . A computer-implemented method of using machine learning to predict dwell times for shipments along routes, the method comprising:
training, by one or more computer processors, a first sub-machine learning model using at least: (i) a first subset of training route data of one or more training routes, and (ii) a first subset of training dwell time data corresponding to the first subset of the training route data, wherein the training route data comprises, for each of the one or more training routes, a set of training stops associated with that training route, and wherein the training dwell time data comprises a plurality of training dwell times, each of which indicates an elapsed time between when a corresponding shipment associated with the training route data arrives at a training stop of the set of training stops and when the corresponding shipment departs the training stop;
training, by the one or more computer processors, a second sub-machine learning model using at least: (i) a second subset of the training route data of the one or more training routes, and (ii) a second subset of the training dwell time data corresponding to the second subset of the training route data, where the first subset of the training dwell time data comprises a first subset of the plurality of training dwell times that are less than a second subset of the plurality of training dwell times included in the second subset of the training dwell time data;
storing the first sub-machine learning model and the second sub-machine learning model in a memory, wherein each of the first sub-machine learning model and the second sub-machine learning model is a gradient-boosted random forest of decision trees;
receiving, by the one or more computer processors from an electronic logging device integrated within a vehicle associated with a shipment of interest, a set of GPS pings corresponding to a set of locations of the vehicle;
in response to receiving the set of GPS pings, automatically determining, by the one or more processors based on the set of GPS pings, a route for the shipment of interest, including detecting, based on a portion of the set of GPS pings indicating that the vehicle is within coordinates of a geofence boundary around a stop along the route, an arrival time to the stop;
selecting, by the one or more computer processors based on the arrival time to the stop, either the first sub-machine learning model or the second sub-machine learning model;
analyzing, by the one or more computer processors using either the first sub-machine learning model or the second sub-machine learning model that was selected, the route data that includes the stop;
based on the analyzing of the route data, outputting, by either the first sub-machine learning model or the second sub-machine learning model that was selected, a predicted dwell time at the stop along the route for the shipment of interest, wherein the predicted dwell time (i) predicts an elapsed time between when the shipment of interest arrives at the stop and when the shipment of interest departs the stop, and (ii) a probability of the elapsed time;
determining, by the one or more computer processors, that the probability of the elapsed time is less than a threshold probability; and
in response to determining that the probability of the elapsed time is less than the threshold probability:
availing, via a visual dashboard, the probability of the elapsed time, wherein the visual dashboard comprises:
a first section, visually distinguished by a first color, representing a first probability range and including the probability of the elapsed time, and
a second section, visually distinguished by a second color different from the first color, representing a second probability range different from the first probability range and including at least one additional probability respectively associated with at least one additional elapsed time and respectively corresponding to at least one additional shipment of interest,
automatically generating, by the one or more computer processors, an electronic communication that indicates a potential intervention associated with the stop, and
automatically sending, to an entity by the one or more computer processors, the electronic communication.
2 . The computer-implemented method of claim 1 , further comprising:
obtaining, at the one or more computer processors, (i) holdout route data of a holdout shipment, and (ii) a holdout dwell time corresponding to the holdout route data;
analyzing, by the one computer processors using either the first sub-machine learning model or second sub-machine learning model that was selected, the holdout route data;
based on the analyzing of the holdout route data, outputting, by either the first sub-machine learning model or the second sub-machine learning model that was selected, a predicted holdout dwell time;
comparing, by the one or more computer processors, the predicted holdout dwell time to the holdout dwell time; and
based on the comparing of the predicted holdout dwell time to the holdout dwell time, adjusting, by the one or more computer processors, one or more weights of either the first sub-machine learning model or the second sub-machine learning model.
3 . The computer-implemented method of claim 1 , further comprising:
receiving, by the one or more computer processors, a location of interest corresponding to the shipment of interest, wherein the location of interest is subsequent to the stop for the shipment of interest; and
determining, by the one or more computer processors, an estimated time of arrival (ETA) for the shipment of interest to the location of interest based on the predicted dwell time at the stop.
4 . The computer-implemented method of claim 3 , wherein the location of interest is an additional stop for the shipment of interest, and wherein the computer-implemented method further comprises:
analyzing, by the one or more computer processors using either the first sub-machine learning model or the second sub-machine learning model that was selected, the ETA for the shipment of interest to the additional stop; and
based on the analyzing of the ETA, outputting, by either the first sub-machine learning model or the second sub-machine learning model that was selected, an additional predicted dwell time at the additional stop for the shipment of interest.
5 . The computer-implemented method of claim 1 , wherein one or both of the training route data or the route data include data associated with two or more of: a shipper, a carrier, a stop location, a planned stop arrival time, an observed stop arrival time, an observed stop departure time, or a total number of stops.
6 . A computer system for using machine learning to predict dwell times for shipments along routes, the method comprising:
one or more processors;
a program memory coupled to the one or more processors and storing executable instructions that, when executed by the one or more processors, cause the computer system to:
train a first sub-machine learning model using at least: (i) a first subset of training route data of one or more training routes, and (ii) a first subset of training dwell time data corresponding to the first subset of training route data, wherein the training route data comprises, for each of the one or more training routes, a set of training stops associated with that training route, and wherein the training dwell time data comprises a plurality of training dwell times, each of which indicates an elapsed time between when a corresponding shipment associated with the training route data arrives at a training stop of the set of training stops and when the corresponding shipment departs the training stop,
train a second sub-machine learning model using at least (i) a second subset of the training route data of the one or more training routes, and (ii) a second subset of the training dwell time data corresponding to the second subset of the training route data, where the first subset of the training dwell time data comprises a first subset of the plurality of training dwell times that are less than a second subset of the plurality of training dwell times included in the second subset of the training dwell time data,
store the first sub-machine learning model and the second sub-machine learning model in the program memory, wherein each of the first sub-machine learning model and the second sub-machine learning model is a gradient-boosted random forest of decision trees,
receive, from an electronic logging device integrated within a vehicle associated with a shipment of interest, a set of GPS pings corresponding to a set of locations of the vehicle,
in response to receiving the set of GPS pings, automatically determine, based on the set of GPS pings, a route for the shipment of interest including detecting, based on a portion of the set of GPS pings indicating that the vehicle is within coordinates of a geofence boundary around a stop along the route, an arrival time to the stop,
select, based on the arrival time to the stop, either the first sub-machine learning model or the second sub-machine learning model,
analyze, using either the first sub-machine learning model or the second sub-machine learning model that was selected, the route data that includes the stop,
based on the analyzing of the route data, output, by either the first sub-machine learning model or the second machine learning model that was selected, a predicted dwell time at the stop along the route for the shipment of interest, wherein the predicted dwell time (i) predicts an elapsed time between when the shipment of interest arrives at the stop and when the shipment of interest departs the stop, and (ii) a probability of the elapsed time,
determine that the probability of the elapsed time is less than a threshold probability, and
in response to determining that the probability of the elapsed time is less than the threshold probability:
avail, via a visual dashboard, the probability of the elapsed time, wherein the visual dashboard comprises:
a first section, visually distinguished by a first color, representing a first probability range and including the probability of the elapsed time, and
a second section, visually distinguished by a second color different from the first color, representing a second probability range different from the first probability range and including at least one additional probability respectively associated with at least one additional elapsed time and respectively corresponding to at least one additional shipment of interest,
automatically generate an electronic communication that indicates a potential intervention associated with the stop, and
automatically send, to an entity, the electronic communication.
7 . The computer system of claim 6 , wherein the executable instructions further cause the computer system to:
obtain (i) holdout route data of a holdout shipment, and (ii) a holdout dwell time corresponding to the holdout route data;
analyze, using either the first sub-machine learning model or second sub-machine learning model that was selected, the holdout route data;
based on the analyzing of the holdout route data, output, by either the first sub-machine learning model or the second sub-machine learning model that was selected, a predicted holdout dwell time;
compare the predicted holdout dwell time to the holdout dwell time; and
based on the comparing of the predicted holdout dwell time to the holdout dwell time, adjust one or more weights of either the first sub-machine learning model or the second sub-machine learning model.
8 . The computer system of claim 6 , wherein one or both of the training route data or the route data include data associated with two or more of: a shipper, a carrier, a stop location, a planned stop arrival time, an observed stop arrival time, an observed stop departure time, or a total number of stops.
9 . The computer system of claim 6 , wherein the executable instructions further cause the computer system to:
receive a location of interest corresponding to the shipment of interest, wherein the location of interest is subsequent to the stop for the shipment of interest; and
determine an estimated time of arrival (ETA) for the shipment of interest to the location of interest based on the predicted dwell time at the stop.
10 . A tangible, non-transitory computer-readable medium storing executable instructions for using machine learning to predict dwell times for shipments along routes that, when executed by one or more processors of a computer system, cause the computer system to:
train a first sub-machine learning model using at least: (i) a first subset of training route data of one or more training routes, (ii) a first subset of training dwell time data corresponding to the first subset of training route data, wherein the training route data comprises, for each of the one or more training routes, a set of training stops associated with that training route, and wherein the training dwell time data comprises a plurality of training dwell times, each of which indicates an elapsed time between when a corresponding shipment associated with the training route data arrives at a training stop of the set of training stops and when the corresponding shipment departs the training stop;
train a second sub-machine learning model using at least: (i) a second subset of the training route data of the one or more training routes, and (ii) a second subset of the training dwell time data corresponding to the second subset of the training route data, where the first subset of the training dwell time data comprises a first subset of the plurality of training dwell times that are less than a second subset of the plurality of training dwell times included in the second subset of the training dwell time data;
store the first sub-machine learning model and second sub-machine learning model in a memory, wherein each of the first sub-machine learning model and the second sub-machine learning model is a gradient-boosted random forest of decision trees;
receive, from an electronic logging device integrated within a vehicle associated with a shipment of interest, a set of GPS pings corresponding to a set of locations of the vehicle;
in response to receiving the set of GPS pings, automatically determine, based on the set of GPS pings, a route for the shipment of interest including detecting, based on a portion of the set of GPS pings indicating that the vehicle is within coordinates of a geofence boundary around a stop along the route, an arrival time to the stop;
select, based on the arrival time to the stop, either the first sub-machine learning model or the second sub-machine learning model;
analyze, using either the first sub-machine learning model or the second sub-machine learning model that was selected, the route data that includes the stop;
based on the analyzing of the route data, output, by either the first sub-machine learning model or the second sub-machine learning model that was selected, a predicted dwell time at the stop along the route for the shipment of interest, wherein the predicted dwell time (i) predicts an elapsed time between when the shipment of interest arrives at the stop and when the shipment of interest departs the stop, and (ii) a probability of the elapsed time;
determine that the probability of the elapsed time is less than a threshold probability; and
in response to determining that the probability of the elapsed time is less than the threshold probability:
avail, via a visual dashboard, the probability of the elapsed time, wherein the visual dashboard comprises:
a first section, visually distinguished by a first color, representing a first probability range and including the probability of the elapsed time, and
a second section, visually distinguished by a second color different from the first color, representing a second probability range different from the first probability range and including at least one additional probability respectively associated with at least one additional elapsed time and respectively corresponding to at least one additional shipment of interest,
automatically generate an electronic communication that indicates a potential intervention associated with the stop, and
automatically send, to an entity, the electronic communication.
11 . The tangible, non-transitory computer-readable medium of claim 10 , wherein the executable instructions further cause the computer system to:
obtain (i) holdout route data of a holdout shipment, and (ii) a holdout dwell time corresponding to the holdout route data;
analyze, using either the first sub-machine learning model or the second sub-machine learning model that was selected, the holdout route data;
based on the analyzing of the holdout route data, output, by either the first sub-machine learning model or the second sub-machine learning model that was selected, a predicted holdout dwell time;
compare the predicted holdout dwell time to the holdout dwell time; and
based on the comparing of the predicted holdout dwell time to the holdout dwell time, adjust one or more weights of either the first sub-machine learning model or the second sub-machine learning model.
12 . The tangible, non-transitory computer-readable medium of claim 10 , wherein the executable instructions further cause the computer system to:
receive a location of interest corresponding to the shipment of interest, wherein the location of interest is subsequent to the stop for the shipment of interest; and
determine an estimated time of arrival (ETA) for the shipment of interest to the location of interest based on the predicted dwell time at the stop.