Trend Attribute Encoding for Time-Series Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated modeling systems rely on static snapshots of attribute data, failing to utilize time series data effectively, leading to inaccurate predictive outputs and inefficient resource consumption.
Innovation Solution
Generating trend attributes through frequency transforms, such as discrete cosine transforms (DCT), to encode time-series information, which are used as inputs for automated modeling algorithms, reducing the need for multiple dataset iterations and optimizing resource usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If static snapshots of attribute data are used for training automated modeling systems, then the system implementation is simple, but predictive accuracy deteriorates due to failure to utilize time series data
Solution Approach 1:
The time series data is segmented into discrete time intervals (e.g., daily, weekly, monthly snapshots), and trend attributes are calculated for each segment. This segmentation allows the system to process time series data in manageable chunks rather than as a continuous stream, improving predictive accuracy while controlling complexity.
Solution Approach 2:
Trend attributes are pre-calculated from historical time series data before being fed into the automated modeling system. This preliminary action of computing trends (e.g., rate of change, acceleration, cyclic patterns) in advance reduces the complexity during actual modeling operations while enhancing predictive accuracy with rich temporal features.
2Measurement precision
If multiple dataset iterations are used to capture time-based characteristics, then predictive accuracy improves, but computational resource consumption increases
Solution Approach 1:
The invention extracts essential time-based characteristics (trends, cyclic patterns, rate of change) from the full time series data and represents them as compact trend attributes. This extraction reduces the data volume significantly while retaining the critical temporal information needed for accurate predictions, thereby lowering computational resource consumption.
Solution Approach 2:
The system transforms time series data from the time domain to the frequency domain using transforms like DCT (Discrete Cosine Transform). This parameter change allows representation of temporal patterns as frequency components, enabling efficient compression and processing while maintaining predictive accuracy.
3Loss of information
If full time series data is processed directly by automated modeling systems, then complete time-based information is utilized, but data processing efficiency deteriorates
Solution Approach 1:
Instead of processing the entire time series data directly, the system creates compressed representations (copies) in the form of trend attributes that capture the essential time-based characteristics. These trend attribute copies retain the critical temporal information while being much more efficient to process by the automated modeling system.
Data Source
Figure 1
Figure 2~4
Figure 5
AI summary
Certain aspects involve updating data structures to indicate relationships between attribute trends and response variables used for training automated modeling systems. For example, a data structure stores data for training an automated modeling algorithm. The training data includes attribute values for multiple entities over a time period. A computing system generates, for each entity, at least one trend attribute that is a function of a respective time series of attribute values. The computing system modifies the data structure to include the generated trend attributes and updates the training data to include trend attribute values for the trend attributes. The computing system trains the automated modeling algorithm with the trend attribute values from the data structure. In some aspects, trend attributes are generated by applying a frequency transform to a time series of attribute values and selecting, as trend attributes, some of the coefficients generated by the frequency transform.