System to predict service level failure in supply chains
A system and method are disclosed for a low-touch centralized system to predict service level failure in a supply chain using machine learning. Embodiments include receiving only historical supply chain data from an archiving system for one or more supply chain entities storing items at stocking locations, predicting one or more supply chain events during a prediction period by applying a predictive model to a sample of historical supply chain data, calculating an occurrence risk score for at least one of the one or more supply chain events and indicating a possibility that the at least one of the one or more supply chain events will occur, generating one or more alerts identifying at least one item and at least one alert stocking location, rendering an alert heatmap visualization comprising one or more selectable user interface elements, and provide one or more tools for initiating corrective actions to be undertaken.
1. A low-touch centralized system to predict service failure in the supply chain using machine learning, comprising:
a server comprising a processor and a memory, the server configured to:
receive only historical supply chain data from an archiving system, the archiving system storing historical supply chain data from a supply chain network comprising one or more supply chain entities, wherein the one or more supply chain entities store one or more items at one or more stocking locations;
aggregate the historical supply chain data to a certain granularity level during a training phase of a gradient boosted trees machine learning model, wherein the aggregated historical supply chain data comprises a sample;
train the gradient boosted trees machine learning model using the sample;
predict one or more supply chain failures during a prediction period by applying a prediction model to a sample of historical supply chain data comprising a sample period earlier than the prediction period; each of the one or more predicted supply chain failures associated with at least one item of the one or more items and at least one stocking location of the one or more stocking locations during a prediction period, wherein the prediction model comprises the gradient boosted trees machine learning model;
calculate an occurrence risk score for at least one of the one or more predicted supply chain failures, the occurrence risk score indicating a possibility that the at least one of the one or more predicted supply chain failures will occur;
determine precision and recall measurements for the prediction model;
determine whether to retrain the gradient boosted trees machine learning model;
calculate a magnitude and a direction of an effect one or more prediction features have on the occurrence risk score;
generate one or more alerts for the one or more supply chain events, each of the one or more alerts representing an output value whose metrics indicate a threshold of greater than 50% chance of occurring associated with at least one alert supply chain event of the one or more supply chain events, the one or more alerts comprising at least one alert item and at least one alert stocking location, the at least one alert item identifying the at least one item of the at least one alert supply chain event and the at least one alert stocking location identifying the at least one stocking location of the at least one alert supply chain event;
provide one or more tools for initiating one or more corrective actions to be undertaken in order to resolve one or more underlying causes of the at least one alert supply chain event; and
initiate the one or more corrective actions by instructing automated machinery to produce one or more products based, at least in part, on the one or more predicted supply chain failures.
2. The system of claim 1 , wherein the server is further configured to:
identify the one or more prediction features relied on by the prediction model to generate the occurrence risk score, the one or more prediction features associated with one or more supply chain features.
3. The system of claim 2 , wherein the server is further configured to:
calculate a precision value of the selected alert for the supply chain event, the precision value indicating a probability the supply chain event will occur based on the past performance of the prediction model for predicting the supply chain event.
4. The system of claim 3 , wherein the prediction model comprises a gradient boosted trees machine learning model, and the prediction period and an amount of time for completing one or more corrective actions are less than a planning horizon.
5. The system of claim 4 , wherein the supply chain event is a service level failure, and the occurrence risk score is a failure risk score.
6. The system of claim 5 , wherein the server is further configured to:
identify one or more properties of the one or more alerts;
receive a filter comprising one or more criteria;
detect each of the one or more alerts having one or properties corresponding to the one or more criteria of the filter; and
filter each of the one or more alerts detected as having one or more properties corresponding to the one or more criteria of the filter.
7. The system of claim 6 , wherein the properties of the one or more alerts comprises the one or more underlying causes of the at least one alert supply chain failure, the one or more underlying causes comprises a production planning problem, and the one or more corrective actions to be taken in order to prevent the one or more underlying causes comprises a time period shorter than a production planning horizon.
8. A method for predicting service failures in the supply chain using machine learning, comprising:
receiving, by a server having a processor and a memory, only historical supply chain data from an archiving system, the archiving system storing historical supply chain data from a supply chain network comprising one or more supply chain entities, wherein the one or more supply chain entities store one or more items at one or more stocking locations;
aggregating, by the server, the historical supply chain data to a certain granularity level during a training phase of a gradient boosted trees machine learning model, wherein the aggregated historical supply chain data comprises a sample;
training, by the server, the gradient boosted trees machine learning model using the sample;
predicting, by the server, one or more supply chain failures during a prediction period by applying a prediction model to a sample of historical supply chain data comprising a sample period earlier than the prediction period; each of the one or more predicted supply chain failures associated with at least one item of the one or more items and at least one stocking location of the one or more stocking locations during a prediction period, wherein the prediction model comprises the gradient boosted trees machine learning model;
calculating, by the server, an occurrence risk score for at least one of the one or more supply chain events, the occurrence risk score indicating a possibility that the at least one of the one or more supply chain events will occur;
determining, by the server, precision and recall measurements for the prediction model;
determining, by the server, whether to retrain the gradient boosted trees machine learning model;
calculating a magnitude and a direction of an effect one or more prediction features have on the occurrence risk score;
generating, by the server, one or more alerts for the one or more supply chain events, each of the one or more alerts representing an output value whose metrics indicate a threshold of greater than 50% chance of occurring associated with at least one alert supply chain event of the one or more supply chain events, the one or more alerts comprising at least one alert item and at least one alert stocking location, the at least one alert item identifying the at least one item of the at least one alert supply chain event and the at least one alert stocking location identifying the at least one stocking location of the at least one alert supply chain event;
providing, by the server, one or more tools for initiating one or more corrective actions to be undertaken in order to resolve one or more underlying causes of the at least one alert supply chain event; and
initiating, by the server, the one or more corrective actions by instructing automated machinery to produce one or more products based, at least in part, on the one or more predicted supply chain failures.
9. The method of claim 8 , further comprising:
identifying, by the server, the one or more prediction features relied on by the prediction model to generate the occurrence risk score, the one or more prediction features associated with one or more supply chain features.
10. The method of claim 9 , further comprising:
calculating, by the server, a precision value of the selected alert for the supply chain event, the precision value indicating a probability the supply chain failure will occur based on the past performance of the prediction model for predicting the supply chain event.
11. The method of claim 10 , wherein the prediction model comprises a gradient boosted trees machine learning model, and the prediction period and an amount of time for completing one or more corrective actions are less than a planning horizon.
12. The method of claim 11 , wherein the supply chain event is a service level failure, and the occurrence risk score is a failure risk score.
13. The method of claim 12 , further comprising:
identifying, by the server, one or more properties of the one or more alerts;
receiving, by the server, a filter comprising one or more criteria;
detecting, by the server, each of the one or more alerts having one or properties corresponding to the one or more criteria of the filter; and
filtering, by the server, each of the one or more alerts detected as having one or more properties corresponding to the one or more criteria of the filter.
14. The method of claim 13 , wherein the properties of the one or more alerts comprises the one or more underlying causes of the at least one alert supply chain event, the one or more underlying causes comprises a production planning problem, and the one or more corrective actions to be taken in order to prevent the one or more underlying causes comprises a time period shorter than a production planning horizon.
15. A non-transitory computer-readable medium embodied with software for a low-touch centralized system to predict service failure in the supply chain using machine learning, the software when executed configured to:
receive only historical supply chain data from an archiving system, the archiving system storing historical supply chain data from a supply chain network comprising one or more supply chain entities, wherein the one or more supply chain entities store one or more items at one or more stocking locations;
aggregate the historical supply chain data to a certain granularity level during a training phase of a gradient boosted trees machine learning model, wherein the aggregated historical supply chain data comprises a sample;
train the gradient boosted trees machine learning model using the sample;
predict one or more supply chain events during a prediction period by applying a prediction model to a sample of historical supply chain data comprising a sample period earlier than the prediction period; each of the one or more supply chain events associated with at least one item of the one or more items and at least one stocking location of the one or more stocking locations during a prediction period, wherein the prediction model comprises the gradient boosted trees machine learning model;
calculate an occurrence risk score for at least one of the one or more supply chain events, the occurrence risk score indicating a possibility that the at least one of the one or more supply chain events will occur;
determine precision and recall measurements for the prediction model;
determine whether to retrain the gradient boosted trees machine learning model;
calculate a magnitude and a direction of an effect one or more prediction features have on the occurrence risk score;
generate one or more alerts for the one or more supply chain events, each of the one or more alerts representing an output value whose metrics indicate a threshold of greater than 50% chance of occurring and associated with at least one alert supply chain event of the one or more supply chain events, the one or more alerts comprising at least one alert item and at least one alert stocking location, the at least one alert item identifying the at least one item of the at least one alert supply chain event and the at least one alert stocking location identifying the at least one stocking location of the at least one alert supply chain event;
provide one or more tools for initiating one or more corrective actions to be undertaken in order to resolve one or more underlying causes of the at least one alert supply chain event; and
initiate the one or more corrective actions by instructing automated machinery to produce one or more products based, at least in part, on the one or more predicted supply chain events.
16. The non-transitory computer-readable medium of claim 15 , wherein the software when executed is further configured to:
identify the one or more prediction features relied on by the prediction model to generate the occurrence risk score, the one or more prediction features associated with one or more supply chain features.
17. The non-transitory computer-readable medium of claim 16 , wherein the software when executed is further configured to:
calculate a precision value of the selected alert for the supply chain event, the precision value indicating a probability the supply chain event will occur based on the past performance of the prediction model for predicting the supply chain event.
18. The non-transitory computer-readable medium of claim 17 , wherein the prediction model comprises a gradient boosted trees machine learning model, and the prediction period and an amount of time for completing one or more corrective actions are less than a planning horizon.
19. The non-transitory computer-readable medium of claim 18 , wherein
the supply chain event is a service level failure, and the occurrence risk score is a failure risk score.
20. The non-transitory computer-readable medium of claim 19 , wherein
the software when executed is further configured to:
identify one or more properties of the one or more alerts;
receive a filter comprising one or more criteria;
detect each of the one or more alerts having one or properties corresponding to the one or more criteria of the filter; and
filter each of the one or more alerts detected as having one or more properties corresponding to the one or more criteria of the filter.