Time-Series Feature Weighting for Explainable Model Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing models, such as machine learning models, often lack transparency regarding feature importance weights, making them less explainable and reducing their predictive accuracy due to the lack of consideration for time-series related information.
Innovation Solution
A proxy model is trained using historical actual values and model outputs to determine time-series aware weights, which are then used to refine the original model, incorporating time-series patterns and enhancing its predictive ability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are trained without historical time series data, then the model structure remains simple and training is faster, but the model lacks accuracy and explainability in real-world applications
Solution Approach 1:
The patent applies preliminary action by pre-processing historical time series data into features before training the proxy model. Historical actual data is transformed into a structured format that captures time-series patterns, allowing the model to learn from past trends without requiring complex real-time processing during inference.
Solution Approach 2:
The patent introduces a proxy model as an intermediary between the original complex model and the historical data. This proxy model serves as a mediator that learns from historical time series data and generates explainable feature importance weights, which are then used to refine the original model without directly training the original model on the complex historical data.
2Loss of information
If feature importance weights are not determined, then the model operates as a black box with unascertainable weights, but determining weights adds computational overhead and training time
Solution Approach 1:
The patent creates a copy of the original model in the form of a proxy model. This proxy model is trained separately on historical data to determine feature importance weights. Once the proxy model generates the explainability metrics, the original model can be refined using these weights without retraining the entire original model from scratch, thus reducing the time loss.
3Reliability
If historical actual data is not incorporated, then the model cannot account for time-series patterns, but incorporating historical data requires additional data processing and storage
Solution Approach 1:
The patent segments the historical time series data into discrete features that can be processed independently. By transforming the continuous historical data into structured feature representations, the system reduces the complexity of processing and storing historical data while maintaining the ability to capture time-series patterns for improved predictive performance.
Data Source
AI summary
A system described herein may receive a set of outputs of a first model, which have been generated by the first model based on a set of inputs, and identify a set of historical values that correspond to the set of inputs and the set of outputs. The inputs and the historical values may be associated with the same time series. The system may train a second model based on the set of inputs to the first model, the set of outputs of the first model, and the set of historical values that correspond to the set of inputs and the set of outputs. The system may determine, based on training the second model, a set of weights associated with the set of historical values; and refine the first model based on the set of weights associated with the set of historical value.