SYSTEM AND METHOD FOR DETERMINING A TRANSIT PREDICTION MODEL
A method for prediction model determination can include: determining a set of models, training each model, determining package transit data, evaluating the set of models, selecting a model from the set of models, predicting package transit data using the selected model, and/or any other suitable element.
1 . A system, comprising:
an interface configured to receive a request for a target package; and
a processing system configured to:
a) determine actual transit times for a set of packages;
b) train a set of models using supervised learning, wherein each model is trained using a different set of historic transit data;
c) for each model:
determine predicted transit times for the set of packages using the model; and
determine individual evaluation metrics for each of a set of time periods based on the predicted and actual transit times for the set of packages;
d) select a model from the set of models based on the individual evaluation metrics for each trained model; and
e) predict a transit time for the target package using the selected model;
wherein the interface returns the predicted transit time for the target package.
2 . The system of claim 1 , wherein each of the set of packages is associated with an evaluation period, wherein each of the set of time periods is within the evaluation period.
3 . The system of claim 1 , wherein each of the set of packages is delivered within the evaluation period.
4 . The system of claim 1 , wherein the processing system is further configured to: for each model, aggregate the individual evaluation metrics to determine an overall evaluation metric, wherein selecting a model from the set of models based on the individual evaluation metrics comprises selecting the model based on the overall evaluation metric for each trained model.
5 . The system of claim 1 , wherein the interface is further configured to return the predicted transit time for the target package, wherein the predicted transit time for the target package is used to select a shipping carrier service.
6 . The system of claim 5 , wherein a label is generated for the selected shipping carrier service, wherein the target package is shipped using the label.
7 . A system, comprising:
a processing system configured to:
a) determine actual transit times for a set of packages;
b) train each of a set of models, wherein each model learns to predict transit times based on historic transit data selected based on a historic training data selection rule associated with the respective model;
c) for each trained model:
determine predicted transit times for the set of packages using the trained model;
determine individual evaluation metrics based on the predicted and actual transit times for the set of packages; and
aggregate the individual evaluation metrics; and
d) select a model from the set of models based on the respective aggregated evaluation metrics, wherein the selected model is used to predict a transit time for a target package.
8 . The system of claim 7 , wherein the individual evaluation metrics comprise individual evaluation metrics for each subperiod of an evaluation period.
9 . The system of claim 8 , wherein the individual evaluation metrics for each subperiod are determined based on the predicted and actual transit times for a subset of packages in the set of packages, wherein each of the subset of packages is delivered within the respective subperiod.
10 . The system of claim 8 , wherein each subperiod comprises a day.
11 . The system of claim 7 , wherein the target package is associated with a shipment created within a prediction period, wherein the selected model is used to predict a transit time for each of a set of target packages created within the prediction period.
12 . The system of claim 11 , wherein the processing system is configured to repeat a)-d) for a successive prediction period, wherein the set of models are retrained for the successive prediction period.
13 . The system of claim 7 , wherein predicting the transit time for the target package comprises:
determining target historic transit data based on the historic training data selection rule associated with the selected model; and
retraining the selected model on the target historic transit data, wherein the transit time for the target package is predicted using the retrained selected model.
14 . The system of claim 7 , wherein each model learns to predict transit times based on a distribution of the selected historic transit data.
15 . The system of claim 14 , wherein each model learns to predict transit times based on a predetermined percentile of the distribution of the selected historic transit data.
16 . The system of claim 7 , wherein the historic transit data is associated with at least one of: a shipping carrier, a shipping carrier service, or a shipping lane.