IP Library › Granted Patent US 12705549
Granted Patent B2
US 12705549 · App. 18/078,276 · Granted Aug 11, 2026

System and method of segmenting data and forecasting by a combination of models trained on segmented data

Inventors: Matt Nitzken (Boston, MA); David McGarry (Boston, MA); Roman Midianyi (Kyiv, UA); Anatolli Stehni (Kyiv, UA)
Assignee: DataRobot, Inc.
G06N20/20
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12705549
App. No.
18/078,276
Granted
Aug 11, 2026
Kind
B2
Abstract

Segmenting data and forecasting by a combination of models trained on segmented data is provided. A system compares, with a first model, values of timestamps corresponding to data points to determine a time series dependency between the data points. The system generates, with the first model and based on the time series dependency, a first cluster with first data points and a second cluster with second data points. The system allocates, by a controller, a second model to the first cluster, and a third model to the second cluster. The system trains the second model based on the time series dependency and the first data points. The system trains the third model based on the time series dependency and the second data points. The system generates a fourth model based on a combination of the second trained model and the third trained model.

Claims (49)

1 . A system, comprising:

a data processing system comprising memory and one or more processors configured to perform operations including:

clustering, with a first model, a plurality of time series data sets represented by a plurality of data points, wherein each data point of the plurality of data points includes a set of one or more feature values characterizing a respective time series data set of the plurality of time series data sets, wherein the clustering is based on the sets of one or more feature values, and wherein the clustering produces at least a first cluster and a second cluster, the first cluster including one or more first time series data sets of the plurality of time series data sets represented by one or more first data points of the plurality of data points, the second cluster including one or more second time series data sets of the plurality of time series data sets represented by one or more second data points of the plurality of data points;

allocating a second model to the first cluster of one or more first time series data sets and a third model to the second cluster of one or more second time series data sets;

training the second model based on the one or more first time series data sets corresponding to the one or more first data points, and training the third model based on the one or more second time series data sets corresponding to the one or more second data points;

generating a fourth model based on a combination of the second trained model and the third trained model;

providing, to the fourth model, a request to generate a forecast value based on one or more time series data points; and

generating, based on input to at least one of the second model or the third model including one or more of the time series data points, an output including a forecast based on the time series data points.

2 . The system of claim 1 , wherein the first model comprises a clustering model, and wherein the operations further include providing, in response to user input, a presentation based on the fourth model, the first data points, and the second data points.

3 . The system of claim 1 , wherein the second model comprises a first supervised model and the third model comprises a second supervised model.

4 . The system of claim 3 , wherein the first supervised model is configured to generate an output based on one or more characteristics of the first cluster.

5 . The system of claim 4 , wherein the second supervised model is configured to generate an output based on one or more characteristics of the second cluster.

6 . The system of claim 1 , wherein the operations further include:

determining, based on one or more of the time series data points, that the time series data points correspond to the second model;

selecting the second model based on the determination that the time series data points correspond to the second model; and

generating, based on input to the second model including one or more of the time series data points, the output including the forecast based on the time series data points.

7 . The system of claim 1 , wherein the operations further include:

determining, based on one or more of the time series data points, that the time series data points correspond to the third model;

selecting the third model based on the determination that the time series data points correspond to the third model; and

generating, based on input to the third model including one or more of the time series data points, the output including the forecast based on the time series data points.

8 . The system of claim 1 , wherein the time series data points correspond to a series having one or more values corresponding to at least one of the first cluster or the second cluster.

9 . A method, comprising:

clustering, by a data processing system and with a first model, a plurality of time series data sets represented by a plurality of data points, wherein each data point of the plurality of data points includes a set of one or more feature values characterizing a respective time series data set of the plurality of time series data sets, wherein the clustering is based on the sets of one or more feature values, and wherein the clustering produces at least a first cluster and a second cluster, the first cluster including one or more first time series data sets of the plurality of time series data sets represented by one or more first data points of the plurality of data points, the second cluster including one or more second time series data sets of the plurality of time series data sets represented by one or more second data points of the plurality of data points;

allocating, by the data processing system, a second model to the first cluster of one or more first time series data sets and a third model to the second cluster of one or more second time series data sets;

training, by the data processing system, the second model based on the one or more first time series data sets corresponding to the one or more first data points, and training the third model based on the one or more second time series data sets correspond to the one or more second data points;

generating, by the data processing system, a fourth model based on a combination of the second trained model and the third trained model;

providing, by the data processing system to the fourth model, a request to generate a forecast value based on one or more time series data points; and

generating, by the data processing system based on input to at least one of the second model or the third model including one or more of the time series data points, an output including a forecast based on the time series data points.

10 . The method of claim 9 , further comprising providing, by the data processing system in response to receiving an indication from a user by a user interface, a presentation based on the fourth model, the first data points, and the second data points, wherein the first model comprises a clustering model.

11 . The method of claim 9 , wherein the second model comprises a first supervised model and the third model comprises a second supervised model.

12 . The method of claim 11 , wherein the first supervised model is configured to generate an output based on one or more characteristics of the first cluster.

13 . The method of claim 12 , wherein the second supervised model is configured to generate an output based on one or more characteristics of the second cluster.

14 . The method of claim 9 , further comprising:

determining, by the data processing system based on one or more of the time series data points, that the time series data points correspond to the second model;

selecting, by the data processing system, the second model based on the determination that the time series data points correspond to the second model; and

generating, by the data processing system, based on input to the second model including one or more of the time series data points, the output including the forecast based on the time series data points.

15 . The method of claim 9 , further comprising:

determining, by the data processing system based on one or more of the time series data points, that the time series data points correspond to the third model;

selecting, by the data processing system, the third model based on the determination that the time series data points correspond to the third model; and

generating, by the data processing system, based on input to the third model including one or more of the time series data points, the output including the forecast based on the time series data points.

16 . The method of claim 9 , wherein the time series data points correspond to a series having one or more values corresponding to at least one of the first cluster or the second cluster.

17 . A non-transitory computer readable medium including one or more instructions stored thereon and executable by a processor to perform operations including:

clustering, with a first model, a plurality of time series data sets represented by a plurality of data points, wherein each data point of the plurality of data points includes a set of one or more feature values characterizing a respective time series data set of the plurality of time series data sets, wherein the clustering is based on the sets of one or more feature values, and wherein the clustering produces at least a first cluster and a second cluster, the first cluster including one or more first time series data sets of the plurality of time series data sets represented by one or more first data points of the plurality of data points, the second cluster including one or more second time series data sets of the plurality of time series data sets represented by one or more second data points of the plurality of data points;

allocating a second model to the first cluster of one or more first time series data sets and a third model to the second cluster of one or more second time series data sets;

training the second model based on the one or more first time series data sets corresponding to the one or more first data points, and training the third model based on the one or more second time series data sets correspond to the one or more second data points;

generating a fourth model based on a combination of the second trained model and the third trained model;

providing, to the fourth model, a request to generate a forecast value based on one or more time series data points; and

generating, based on input to at least one of the second model or the third model

including one or more of the time series data points, an output including a forecast based on the time series data points.