# US 12,009,107 B2 · Seasonally adjusted predictive data analysis > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,009,107 B2 - Document ID: US12009107B2 - Status: Granted - Kind: B2 - Application number: 17/015,531 - Filed: 2020-09-09 - Published: 2022-03-10 - Granted: 2024-06-11 - Earliest claimed: 2020-09-09 - Assignees: Optum, Inc. - Applicants: Optum, Inc. - Inventors: Siddharth Garg; Jyoti Nahata; Madhuri Yadav; Danita Kiser; Prashant Singh; Aishwarya Aishwarya - Web page: https://patentgrep.com/patent/US12009107B2 ## CPC - G16H 50/80: for detecting, monitoring or modelling epidemics or pandemics, e.g. flu - G06N 5/04: Inference or reasoning models - G06N 20/00: Machine learning - G16H 50/20: for computer-aided diagnosis, e.g. based on medical expert systems - G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract There is a need for more effective and efficient seasonally-adjusted predictive data analysis solutions. This need can be addressed by, for example, solutions for performing seasonally-adjusted predictive data analysis that use autoregressive integrated moving average (ARIMA) machine learning models. In one example, a method includes: identifying a seasonally-adjusted training input timeseries data object and a seasonally-adjusted testing timeseries data object; generating a trained ARIMA machine learning model using the seasonally-adjusted training input timeseries data object; determining a validation determination for the trained ARIMA machine learning model based on the seasonally-adjusted testing timeseries data object; determining whether the validation determination describes a positive validation; and in response to determining that the validation determination describes the positive validation, enabling performance of a predictive inference using the trained ARIMA machine learning model in order to generate one or more prospective time-lagged predictions and to perform prediction-based actions based on the prospective time-lagged predictions. ## Claims ### Claim 1 A computer-implemented method comprising: identifying, by one or more processors, a seasonally-adjusted training input timeseries data object and a seasonally-adjusted testing timeseries data object for a superior time period within a monitored period, wherein the seasonally-adjusted training input timeseries data object is identified by replacing an outlier value corresponding to an inferior time period within the superior time period with an average value across one or more preceding inferior time periods corresponding to the inferior time period and preceding the superior time period; updating, by the one or more processors, a trained autoregressive integrated moving average (ARIMA) machine learning model using the seasonally-adjusted training input timeseries data object by: determining an optimal differencing order hyper-parameter of the trained ARIMA machine learning model that is configured to transform the seasonally-adjusted training input timeseries data object into a stationary seasonally-adjusted training input timeseries data object, determining an optimal lag order hyper-parameter of the trained ARIMA machine learning model and an optimal moving average order hyper-parameter of the trained ARIMA machine learning model that are collectively configured to optimize a machine learning model utility measure for the trained ARIMA machine learning model, and determining one or more optimal parameters of the trained ARIMA machine learning model that are configured to enable the trained ARIMA machine learning model to describe the seasonally-adjusted training input timeseries data object given the optimal differencing order hyper-parameter, the optimal lag order hyper-parameter, and the optimal moving average order hyper-parameter; determining a validation determination for the trained ARIMA machine learning model based at least in part on the seasonally-adjusted testing timeseries data object; and in response to the validation determination satisfying an accuracy measure threshold, enabling, by the one or more processors, performance of a predictive inference using the trained ARIMA machine learning model in order to generate one or more prospective time-lagged predictions and to perform one or more prediction-based actions based at least in part on the one or more prospective time-lagged predictions. ### Claim 2 (depends on 1) The computer-implemented method of claim 1, wherein identifying the seasonally-adjusted training input timeseries data object comprises: for each training time period of a plurality of training time periods in a training window, determining a raw training input value of a plurality of raw training input values associated with a raw training input timeseries data object; performing seasonal averaging outlier replacement on the plurality of raw training input values to generate a plurality of seasonally averaged training input values; performing logarithmic variance stabilization on the plurality of seasonally averaged training input values in order to generate a transformed training input timeseries data object; determining a trend component, a seasonality component, and an error component for the raw training input timeseries data object; and determining the seasonally-adjusted training input timeseries data object based at least in part on the trend component and the error component. ### Claim 3 (depends on 1) The computer-implemented method of claim 1, wherein identifying the seasonally-adjusted testing timeseries data object comprises: for each testing time period of a plurality of testing time periods in a testing window, determining a raw testing input value of a plurality of raw testing input values associated with a raw testing timeseries data object; performing seasonal averaging outlier replacement on the plurality of raw testing input values to generate a plurality of seasonally averaged testing input values; performing logarithmic variance stabilization on the plurality of seasonally averaged testing input values in order to generate a transformed testing timeseries data object; determining a trend component, a seasonality component, and an error component for the raw testing timeseries data object; and determining the seasonally-adjusted testing timeseries data object based at least in part on the trend component and the error component. ### Claim 4 (depends on 1) The computer-implemented method of claim 1, wherein the machine learning model utility measure is determined based at least in part on at least one of an Akaike information criterion measure, a corrected Akaike information criterion measure, and a Bayesian information criterion measure. ### Claim 5 (depends on 1) The computer-implemented method of claim 1, wherein the one or more prospective time-lagged predictions are determined based at least in part on inferred covariates of the trained ARIMA machine learning model. ### Claim 6 (depends on 1) The computer-implemented method of claim 1, wherein the one or more prospective time-lagged predictions comprise at least one of a prospective one-week-lagged prediction, a prospective two-week-lagged prediction, and a prospective three-week-lagged prediction. ### Claim 7 (depends on 1) The computer-implemented method of claim 1, wherein the trained ARIMA machine learning model is a trained seasonal ARIMA machine learning model. ### Claim 8 (depends on 1) The computer-implemented method of claim 1, wherein determining the validation determination comprises: identifying a plurality of training periods associated with the seasonally-adjusted testing timeseries data object, wherein the plurality of training periods are associated with a training period order; for each subset of the seasonally-adjusted testing timeseries data object that is associated with an ordered portion of the plurality of training periods, determining an inferred prediction for a subsequent training period of the plurality of training periods by processing the subset of the seasonally-adjusted testing timeseries data object using the trained ARIMA machine learning model; and determining the validation determination based at least in part on the seasonally-adjusted testing timeseries data object and each inferred prediction determined based at least in part on a subset of the seasonally-adjusted testing timeseries data object. ### Claim 9 (depends on 1) The computer-implemented method of claim 1, wherein the one or more prospective time-lagged predictions describe one or more influenza outbreak predictions. ### Claim 10 (depends on 9) The computer-implemented method of claim 9, wherein performing the one or more prediction-based actions comprises generating an influenza outbreak prediction user interface object for one or more prospective time-lagged periods based at least in part on the one or more influenza outbreak predictions. ### Claim 11 A system comprising a memory and one or more processors configured: identify a seasonally-adjusted training input timeseries data object and a seasonally-adjusted testing timeseries data object for a superior time period within a monitored period, wherein the seasonally-adjusted training input timeseries data object is identified by replacing an outlier value corresponding to an inferior time period within the superior time period with an average value across one or more preceding inferior time periods corresponding to the inferior time period and preceding the superior time period; update a trained autoregressive integrated moving average (ARIMA) machine learning model using the seasonally-adjusted training input timeseries data object by: determining an optimal differencing order hyper-parameter of the trained ARIMA machine learning model that is configured to transform the seasonally-adjusted training input timeseries data object into a stationary seasonally-adjusted training input timeseries data object, determining an optimal lag order hyper-parameter of the trained ARIMA machine learning model and an optimal moving average order hyper-parameter of the trained ARIMA machine learning model that are collectively configured to optimize a machine learning model utility measure for the trained ARIMA machine learning model, and determining one or more optimal parameters of the trained ARIMA machine learning model that are configured to enable the trained ARIMA machine learning model to describe the seasonally-adjusted training input timeseries data object given the optimal differencing order hyper-parameter, the optimal lag order hyper-parameter, and the optimal moving average order hyper-parameter; determine a validation determination for the trained ARIMA machine learning model based at least in part on the seasonally-adjusted testing timeseries data object; and in response to the validation determination satisfying an accuracy measure threshold, enable performance of a predictive inference using the trained ARIMA machine learning model in order to generate one or more prospective time-lagged predictions and to perform one or more prediction-based actions based at least in part on the one or more prospective time-lagged predictions. ### Claim 12 (depends on 11) The system of claim 11, wherein identifying the seasonally-adjusted training input timeseries data object comprises: for each training time period of a plurality of training time periods in a training window, determining a raw training input value of a plurality of raw training input values associated with a raw training input timeseries data object; performing seasonal averaging outlier replacement on the plurality of raw training input values to generate a plurality of seasonally averaged training input values; performing logarithmic variance stabilization on the plurality of seasonally averaged training input values in order to generate a transformed training input timeseries data object; determining a trend component, a seasonality component, and an error component for the raw training input timeseries data object; and determining the seasonally-adjusted training input timeseries data object based at least in part on the trend component and the error component. ### Claim 13 (depends on 11) The system of claim 11, wherein identifying the seasonally-adjusted testing timeseries data object comprises: for each testing time period of a plurality of testing time periods in a testing window, determining a raw testing input value of a plurality of raw testing input values associated with a raw testing timeseries data object; performing seasonal averaging outlier replacement on the plurality of raw testing input values to generate a plurality of seasonally averaged testing input values; performing logarithmic variance stabilization on the plurality of seasonally averaged testing input values in order to generate a transformed testing timeseries data object; determining a trend component, a seasonality component, and an error component for the raw testing timeseries data object; and determining the seasonally-adjusted testing timeseries data object based at least in part on the trend component and the error component. ### Claim 14 (depends on 11) The system of claim 11, wherein determining the validation determination comprises: identifying a plurality of training periods associated with the seasonally-adjusted testing timeseries data object, wherein the plurality of training periods are associated with a training period order; for each subset of the seasonally-adjusted testing timeseries data object that is associated with an ordered portion of the plurality of training periods, determining an inferred prediction for a subsequent training period of the plurality of training periods by processing the subset of the seasonally-adjusted testing timeseries data object using the trained ARIMA machine learning model; and determining the validation determination based at least in part on the seasonally-adjusted testing timeseries data object and each inferred prediction determined based at least in part on a subset of the seasonally-adjusted testing timeseries data object. ### Claim 15 (depends on 11) The system of claim 11, wherein the one or more prospective time-lagged predictions describe one or more influenza outbreak predictions. ### Claim 16 At least one non-transitory computer-readable storage media including instructions that, when executed by one or more processors, cause the one or more processors to: identify a seasonally-adjusted training input timeseries data object and a seasonally-adjusted testing timeseries data object for a superior time period within a monitored period, wherein the seasonally-adjusted training input timeseries data object is identified by replacing an outlier value corresponding to an inferior time period within the superior time period with an average value across one or more preceding inferior time periods corresponding to the inferior time period and preceding the superior time period; update a trained autoregressive integrated moving average (ARIMA) machine learning model using the seasonally-adjusted training input timeseries data object by: determining an optimal differencing order hyper-parameter of the trained ARIMA machine learning model that is configured to transform the seasonally-adjusted training input timeseries data object into a stationary seasonally-adjusted training input timeseries data object, determining an optimal lag order hyper-parameter of the trained ARIMA machine learning model and an optimal moving average order hyper-parameter of the trained ARIMA machine learning model that are collectively configured to optimize a machine learning model utility measure for the trained ARIMA machine learning model, and determining one or more optimal parameters of the trained ARIMA machine learning model that are configured to enable the trained ARIMA machine learning model to describe the seasonally-adjusted training input timeseries data object given the optimal differencing order hyper-parameter, the optimal lag order hyper-parameter, and the optimal moving average order hyper-parameter; determine a validation determination for the trained ARIMA machine learning model based at least in part on the seasonally-adjusted testing timeseries data object; and in response to the validation determination satisfying an accuracy measure threshold, enable performance of a predictive inference using the trained ARIMA machine learning model in order to generate one or more prospective time-lagged predictions and to perform one or more prediction-based actions based at least in part on the one or more prospective time-lagged predictions. ### Claim 17 (depends on 16) The at least one non-transitory computer-readable storage media of claim 16, wherein identifying the seasonally-adjusted training input timeseries data object comprises: for each training time period of a plurality of training time periods in a training window, determining a raw training input value of a plurality of raw training input values associated with a raw training input timeseries data object; performing seasonal averaging outlier replacement on the plurality of raw training input values to generate a plurality of seasonally averaged training input values; performing logarithmic variance stabilization on the plurality of seasonally averaged training input values in order to generate a transformed training input timeseries data object; determining a trend component, a seasonality component, and an error component for the raw training input timeseries data object; and determining the seasonally-adjusted training input timeseries data object based at least in part on the trend component and the error component. ### Claim 18 (depends on 16) The at least one non-transitory computer-readable storage media of claim 16, wherein identifying the seasonally-adjusted testing timeseries data object comprises: for each testing time period of a plurality of testing time periods in a testing window, determining a raw testing input value of a plurality of raw testing input values associated with a raw testing timeseries data object; performing seasonal averaging outlier replacement on the plurality of raw testing input values to generate a plurality of seasonally averaged testing input values; performing logarithmic variance stabilization on the plurality of seasonally averaged testing input values in order to generate a transformed testing timeseries data object; determining a trend component, a seasonality component, and an error component for the raw testing timeseries data object; and determining the seasonally-adjusted testing timeseries data object based at least in part on the trend component and the error component. ### Claim 19 (depends on 16) The at least one non-transitory computer-readable storage media of claim 16, wherein determining the validation determination comprises: identifying a plurality of training periods associated with the seasonally-adjusted testing timeseries data object, wherein the plurality of training periods are associated with a training period order; for each subset of the seasonally-adjusted testing timeseries data object that is associated with an ordered portion of the plurality of training periods, determining an inferred prediction for a subsequent training period of the plurality of training periods by processing the subset of the seasonally-adjusted testing timeseries data object using the trained ARIMA machine learning model; and determining the validation determination based at least in part on the seasonally-adjusted testing timeseries data object and each inferred prediction determined based at least in part on a subset of the seasonally-adjusted testing timeseries data object. ### Claim 20 (depends on 16) The at least one non-transitory computer-readable storage media of claim 16, wherein the one or more prospective time-lagged predictions describe one or more influenza outbreak predictions.