Forecasting events by modeling time series data
An event forecasting system can filer time series data by identifying outliers and/or data instances that align with an exception, and either remove or replace them. The event forecasting system can further select a set of models expected to be best predictive for the time series data by performing an initial ranking using a first part of the time series data, validated by a second part; and performing a second ranking of the best models using the first part and second parts of the time series data, validated by a third part. Finally, the event forecasting system can output results to a user interface where a user can view the known and predicted data values and make injects to manually adjust the predicted values as desired.
1 . A method for making predictions based on time series data using one or more automatically selected prediction models, the method comprising:
traversing a directed graph that comprises nodes corresponding to simulations;
selecting nodes of the directed graph,
wherein the directed graph specifies dependencies between the simulations such that a particular simulation, corresponding to a particular node with one or more incoming edges, is to be selected for performance when one or more previous simulations, corresponding to one or more previous nodes connected to the one or more incoming edges for the particular node, are complete, and
wherein, for each selected node, a simulation corresponding to the selected node is performed by:
obtaining a time series data set for the corresponding simulation;
automatically selecting one or more top performing models, from a set of available prediction models, for the time series data set by:
dividing the time series data into three sections;
applying each of the available prediction models to a first of the three sections to generate first predictions for each model;
determining a first accuracy score for each of the available prediction models based on a comparison of the first predictions from each of the available prediction models to values in a second of the three sections;
identifying multiple intermediate top ranked models based on the first accuracy scores;
applying each of the multiple intermediate top ranked models to at least the second of the three sections to generate second predictions for each of the multiple intermediate top ranked models;
determining a second accuracy score for each of the multiple intermediate top ranked models based on a comparison of the second predictions from each of the intermediate top ranked models to values in a third of the three sections of the time series data; and
selecting the one or more top performing models based on the second accuracy scores;
generating multiple predictions for the corresponding simulation by applying the selected one or more top performing models to a version of the time series data set;
causing a user interface (UI) to be displayed including a representation of at least some of the multiple predictions generated for the corresponding simulations; and
causing the multiple predictions to be applied in a future-focused application.
2 . The method of claim 1 , wherein a first of the time series data sets for a least one of the corresponding simulations is validated by removing or replacing one or more items in the first time series data set that is identified to be one or more of:
outside a standard deviation of the entire first time series data set;
outside a standard deviation of the data items, from the first time series data set, in a time slot corresponding to that data item;
associated with a user-defined external irregular factor; or
a combination thereof.
3 . The method of claim 1 , wherein the user interface includes at least one graph with sections having different visual characteristics for the time series data sets and the multiple predictions.
4 . The method of claim 1 , wherein the user interface includes at least one graph with sections having different visual characteristics for A) a first section illustrating some of the multiple predictions that can be modified by user injects and B) a second section illustrating some of the multiple predictions that cannot be modified by user injects.
5 . The method of claim 1 , wherein the user interface includes at least one graph with sections having different visual characteristics for A) a first portion illustrating some of the multiple predictions and B) a second portion illustrating user injects.
6 . The method of claim 1 , wherein the identifying the multiple intermediate top ranked models comprises:
selecting the available prediction models with first accuracy scores above a threshold; and/or
selecting a defined number of the available prediction models with the highest first accuracy scores.
7 . A computer-readable storage medium storing instructions that, when executed by a computing system, cause the computing system to perform a process for making predictions based on time series data using one or more automatically selected prediction models, the process comprising:
traversing a directed graph that comprises nodes corresponding to simulations;
selecting nodes of the directed graph,
wherein the directed graph specifies dependencies between the simulations such that a particular simulation, corresponding to a particular node with one or more incoming edges, is to be selected for performance when one or more previous simulations, corresponding to one or more previous nodes connected to the one or more incoming edges for the particular node, are complete, and
wherein, for each selected node, a simulation corresponding to the selected node is performed by:
obtaining a time series data set for the corresponding simulation;
automatically selecting one or more top performing models, from a set of available prediction models, for the time series data set by:
dividing the time series data into multiple sections;
applying each of the available prediction models to a first of the multiple sections to generate first predictions for each model;
determining a first accuracy score for each of the available prediction models based on a comparison of the first predictions from each of the available prediction models to the values in a second of the multiple sections; and
selecting the one or more top performing models based, at least in part, on the first accuracy scores;
generating multiple predictions for the corresponding simulation by applying the selected one or more top performing models to a version of the time series data set; and
causing a user interface (UI) to be displayed including a representation of at least some of the multiple predictions generated for the corresponding simulations.
8 . The computer-readable storage medium of claim 7 , wherein a first of the time series data sets for a least one of the corresponding simulations is validated by removing or replacing one or more items in the first time series data set that is identified to be one or more of:
outside a standard deviation of the entire first time series data set;
outside a standard deviation of the data items, from the first time series data set, in a time slot corresponding to that data item;
associated with a user-defined external irregular factor; or
a combination thereof.
9 . The computer-readable storage medium of claim 7 , wherein the user interface includes at least one graph with sections having different visual characteristics for the time series data sets and the multiple predictions.
10 . The computer-readable storage medium of claim 7 , wherein the user interface includes at least one graph with sections having different visual characteristics for A) a first section illustrating some of the multiple predictions that can be modified by user injects and B) a second section illustrating some of the multiple predictions that cannot be modified by user injects.
11 . The computer-readable storage medium of claim 7 , wherein the user interface includes at least one graph with sections having different visual characteristics for A) a first portion illustrating some of the multiple predictions and B) a second portion illustrating user injects.
12 . The computer-readable storage medium of claim 7 , wherein the selecting the one or more top performing models comprises:
selecting the available prediction models with first accuracy scores above a threshold; and/or
selecting a defined number of the available prediction models with the highest first accuracy scores.
13 . A computing system for making predictions based on time series data using one or more automatically selected prediction models, the computing system comprising:
one or more processors; and
one or more memories storing instructions that, when executed by the one or more processors, cause the computing system to perform a process comprising:
traversing a directed graph that comprises nodes corresponding to simulations;
selecting nodes of the directed graph,
wherein the directed graph specifies dependencies between the simulations such that a particular simulation, corresponding to a particular node with one or more incoming edges, is to be selected for performance when one or more previous simulations, corresponding to one or more previous nodes connected to the one or more incoming edges for the particular node, are complete, and
wherein, for each selected node, a simulation corresponding to the selected node is performed by:
obtaining a time series data set for the corresponding simulation;
automatically selecting one or more top performing models, from a set of available prediction models, for the time series data set by:
dividing the time series data into multiple sections;
applying each of the available prediction models to a first of the multiple sections to generate first predictions for each model;
determining a first accuracy score for each of the available prediction models based on a comparison of the first predictions from each of the available prediction models to the values in a second of the multiple sections; and
selecting the one or more top performing models based, at least in part, on the first accuracy scores;
generating multiple predictions for the corresponding simulation by applying the selected one or more top performing models to a version of the time series data set; and
causing a user interface (UI) to be displayed including a representation of at least some of the multiple predictions generated for the corresponding simulations.
14 . The computing system of claim 13 , wherein a first of the time series data sets for a least one of the corresponding simulations is validated by removing or replacing one or more items in the first time series data set that is identified to be one or more of:
outside a standard deviation of the entire first time series data set;
outside a standard deviation of the data items, from the first time series data set, in a time slot corresponding to that data item;
associated with a user-defined external irregular factor; or
a combination thereof.
15 . The computing system of claim 13 , wherein the user interface includes at least one graph with sections having different visual characteristics for the time series data sets and the multiple predictions.
16 . The computing system of claim 13 , wherein the user interface includes at least one graph with sections having different visual characteristics for A) a first portion illustrating some of the multiple predictions and B) a second portion illustrating user injects.
17 . The computing system of claim 13 , wherein the selecting the one or more top performing models comprises:
selecting the available prediction models with first accuracy scores above a threshold; and/or
selecting a defined number of the available prediction models with the highest first accuracy scores.
18 . The method of claim 1 , wherein a first of the time series data sets for a least one of the corresponding simulations is validated by removing or replacing one or more items in the first time series data set that is identified to be outside a standard deviation of one or more other data items, from the first time series data set, in a same time slot corresponding to the one or more items.
19 . The computer-readable storage medium of claim 7 , wherein a first of the time series data sets for a least one of the corresponding simulations is validated by removing or replacing one or more items in the first time series data set that is identified to be outside a standard deviation of one or more other data items, from the first time series data set, in a same time slot corresponding to the one or more items.
20 . The system of claim 13 , wherein a first of the time series data sets for a least one of the corresponding simulations is validated by removing or replacing one or more items in the first time series data set that is identified to be outside a standard deviation of one or more other data items, from the first time series data set, in a same time slot corresponding to the one or more items.