Analysis and correction of supply chain design through machine learning
A dynamic supply chain planning system for analysis of historical lead time data that uses machine learning algorithms to forecast future lead times based on historical lead time data, weather data and financial data related to locations and dates within the supply chain.
1. A computer-implemented method comprising:
receiving, by a processor, historical lead time data from a database;
extracting, by the processor, weather data from a weather database, the weather data based on a plurality of source locations, a plurality of destinations locations and a plurality of shipment dates, the plurality of source locations, the plurality of destinations locations and the plurality of shipment dates related to the historical lead time data;
extracting, by the processor, economic indicators data from an economic database, the economic indicators data based on the plurality of source locations, the plurality of destinations locations and the plurality of shipment dates;
generating, by the processor, processed historical lead time data by at least one of:
selecting one or more historical lead time features; and
removing outlier data from the historical lead time data;
generating, by the processor, processed weather data by selection of one or more weather features;
generating, by the processor, processed economic indicators data by selection of one or more economic indicators features;
constructing, by the processor, a times series data based on the processed historical lead time data, the processed weather data and the processed economic indicators data;
dividing, by the processor, the time series data into a first subset and a second subset;
training a plurality of machine learning forecasting algorithms on the first subset of the time series data;
testing, each of the plurality of machine learning forecasting algorithms on the second subset of the time series data and selecting, based on the testing, a machine learning forecasting algorithm having a highest accuracy;
retraining the selected machine learning forecasting algorithm based on the time-series data;
generating, by the selected machine learning forecasting algorithm, one or more lead time identifiers;
augmenting, by the processor, the time series data with one or more features, the time series data having a dimensionality greater than a 2-dimensional space and each of the one or more features having a dimensionality orthogonal to the dimensionality of the time series data;
projecting the time series data to 2-dimensional space and reducing a noise thereof;
separating, by the processor, the time series data into one or more groups based on a time density of data points;
selecting, by the processor, an optimal number of clusters for each of the one or more groups;
performing, by the processor, using a machine learning clustering algorithm, clustering on each of the one or more groups to provide a plurality of clusters;
receiving, by the processor, from a user via a user interface, tolerance criteria for separation of the time series data into a plurality of tolerance zones;
separating, by the processor, the time series data into the plurality of tolerance zones;
separating, by the processor, the plurality of clusters in accordance with a tolerance zone of each group;
further separating, by the processor, the plurality of clusters according to the one or more lead time identifiers, to provide one or more further separated clusters; and
automatically adjusting, by the processor, planned lead times for future purchase orders based on the further separated clusters.
2. The computer-implemented method of claim 1 , wherein the one or more features relate to linearity and seasonality.
3. The computer-implemented method of claim 2 , wherein the features are seasonality, upward linearity, flat linearity, and downward linearity.
4. The computer-implemented method of claim 1 , wherein the number of groups is three.
5. The computer-implemented method of claim 1 , further comprising:
pruning, by the processor, each cluster;
dividing, by the processor, each cluster into a plurality of subclusters.
6. A system comprising:
a processor; and
a memory storing instructions that, when executed by the processor, configure the system to:
receive, by the processor, historical lead time data from a database;
extract, by the processor, weather data from a weather database, the weather data based on a plurality of source locations, a plurality of destinations locations and a plurality of shipment dates, the plurality of source locations, the plurality of destinations locations and the plurality of shipment dates related to the historical lead time data;
extract, by the processor, economic indicators data from an economic database, the economic indicators data based on the plurality of source locations, the plurality of destinations locations and the plurality of shipment dates;
generate, by the processor, processed historical lead time data by at least one of:
selecting one or more historical lead time features; and
removing outlier data from the historical lead time data;
generate, by the processor, processed weather data by selection of one or more weather features;
generate, by the processor, processed economic indicators data by selection of one or more economic indicators features;
construct, by the processor, a times series data based on the processed historical lead time data, the processed weather data and the processed economic indicators data;
divide, by the processor, the time series data into a first subset and a second subset;
train a plurality of machine learning forecasting algorithms on the first subset of the time series data;
test, each of the plurality of machine learning forecasting algorithms on the second subset of the time series data and selecting, based on the testing, a machine learning forecasting algorithm having a highest accuracy;
retrain the selected machine learning forecasting algorithm based on the time-series data;
generate, by the selected machine learning forecasting algorithm, one or more lead time identifiers;
augment, by the processor, the time series data with one or more features, the time series data having a dimensionality greater than a 2-dimensional space and each of the one or more features having a dimensionality orthogonal to the dimensionality of the time series data;
project the time series data to 2-dimensional space and reducing a noise thereof;
separate, by the processor, the time series data into one or more groups based on a time density of data points;
select, by the processor, an optimal number of clusters for each of the one or more groups;
perform, by the processor, using a machine learning clustering algorithm, clustering on each of the one or more groups to provide a plurality of clusters;
receive, by the processor, from a user via a user interface, tolerance criteria for separation of the time series data into a plurality of tolerance zones;
separate, by the processor, the time series data into the plurality of tolerance zones;
separate, by the processor, the plurality of clusters in accordance with a tolerance zone of each group;
further separate, by the processor, the plurality of clusters according to the one or more lead time identifiers to provide one or more further separated clusters; and
automatically adjust, by the processor, planned lead times for future purchase orders based on the further separated.
7. The system of claim 6 , wherein the one or more features relate to linearity and seasonality.
8. The system of claim 7 , wherein the features are seasonality, upward linearity, flat linearity, and downward linearity.
9. The system of claim 6 , wherein the number of groups is three.
10. The system of claim 6 , wherein the instructions further configure the system to:
prune, by the processor, each cluster;
divide, by the processor, each cluster into a plurality of subclusters.
11. A non-transitory computer-readable storage medium, the computer-readable storage medium including instructions that when executed by a processor, cause the processor to:
receive, by a processor, historical lead time data from a database;
extract, by the processor, weather data from a weather database, the weather data based on a plurality of source locations, a plurality of destinations locations and a plurality of shipment dates, the plurality of source locations, the plurality of destinations locations and the plurality of shipment dates related to the historical lead time data;
extract, by the processor, economic indicators data from an economic database, the economic indicators data based on the plurality of source locations, the plurality of destinations locations and the plurality of shipment dates;
generate, by the processor, processed historical lead time data by at least one of:
selecting one or more historical lead time features; and
removing outlier data from the historical lead time data;
generate, by the processor, processed weather data by selection of one or more weather features;
generate, by the processor, processed economic indicators data by selection of one or more economic indicators features;
construct, by the processor, a times series data based on the processed historical lead time data, the processed weather data and the processed economic indicators data;
divide, by the processor, the time series data into a first subset and a second subset;
train a plurality of machine learning forecasting algorithms on the first subset of the time series data;
test, each of the plurality of machine learning forecasting algorithms on the second subset of the time series data and selecting, based on the testing, a machine learning forecasting algorithm having a highest accuracy;
retrain the selected machine learning forecasting algorithm based on the time-series data;
generate, by the selected machine learning forecasting algorithm, one or more lead time identifiers;
augment, by the processor, the time series data with one or more features, the time series data having a dimensionality greater than a 2-dimensional space and each of the one or more features having a dimensionality orthogonal to the dimensionality of the time series data;
project the time series data to 2-dimensional space and reducing a noise thereof;
separate, by the processor, the time series data into one or more groups based on a time density of data points;
select, by the processor, an optimal number of clusters for each of the one or more groups;
perform, by the processor, using a machine learning clustering algorithm, clustering on each of the one or more groups to provide a plurality of clusters;
receive, by the processor, from a user via a user interface, tolerance criteria for separation of the time series data into a plurality of tolerance zones;
separate, by the processor, the time series data into the plurality of tolerance zones;
separate, by the processor, the plurality of clusters in accordance with a tolerance zone of each group;
further separate, by the processor, the plurality of clusters according to the one or more lead time identifiers to provide one or more further separated clusters; and
automatically adjust, by the processor, planned lead times for future purchase orders based on the further separated clusters.
12. The non-transitory computer-readable storage medium of claim 11 , wherein the one or more features relate to linearity and seasonality.
13. The non-transitory computer-readable storage medium of claim 12 , wherein the features are seasonality, upward linearity, flat linearity, and downward linearity.
14. The non-transitory computer-readable storage medium of claim 11 , wherein the number of groups is three.
15. The non-transitory computer-readable storage medium of claim 11 , wherein the instructions further include instructions that when executed by the processor, cause the processor to:
prune, by the processor, each cluster;
divide, by the processor, each cluster into a plurality of subclusters.