Time-Series Feature Generation and Validation via Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems lack the ability to generate and validate time-series model outputs in real-time with high granularity, limiting their effectiveness in predicting instantaneous values and identifying contributing factors.
Innovation Solution
A system that segments time-series data into discrete segments, generates models based on input features, and transforms features based on these models to provide real-time outputs and validate model accuracy across multiple time segments, using machine learning techniques to identify contributing factors to instantaneous values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If time-series data is analyzed as a whole, then model validation is simplified, but prediction accuracy for instantaneous values deteriorates
Solution Approach 1:
The patent divides the time-series data into multiple discrete time segments (e.g., hourly, daily, weekly segments) and trains separate models for each segment. This segmentation enables high-granularity prediction of instantaneous values while maintaining manageable model validation through systematic comparison across segments, resolving the contradiction between prediction accuracy and validation complexity.
2Speed
If real-time feature generation is implemented, then responsiveness to current values improves, but computational complexity increases
Solution Approach 1:
The patent pre-trains models on historical time-series data for each discrete time segment before real-time operation. During real-time execution, the system only needs to select the appropriate pre-trained model based on the current time segment and generate features using that model, significantly reducing real-time computational complexity while maintaining high responsiveness.
3Measurement precision
If high-granularity time segments are used, then identification of contributing factors improves, but data processing time increases
Solution Approach 1:
The patent segments time-series data into discrete time segments and trains dedicated models for each segment, enabling precise identification of contributing factors specific to each time granularity. The system processes only the relevant segment data for each prediction task, avoiding unnecessary processing of entire time-series datasets and thus reducing overall data processing time despite high granularity requirements.
Data Source
AI summary
This disclosure relates generally to using machine learning models to generate current time-series features using machine learning and validate time-series machine learning model output. At least one aspect is directed to a system with one or more processors, coupled to memory, to segment a time series range into a first segment for an instance of time, the segment associated with a value for a target feature and a timestamp for the value, segment the time series range into an input segment associated with a plurality of input features and a segment timestamp less than or equal to the timestamp, generate a model trained with input comprising values for the target feature and timestamps for the values less than or equal to the segment timestamp, and transform at least one of the input features based at least on the model.