Causal Error Analysis for Time-Series Predictive Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing predictive models, particularly those using machine learning, suffer from a 'black-box' nature, making it difficult to explain why predicted values deviate from actual values, especially in socially influential infrastructure services, and existing techniques fail to consider causality among variables when determining the cause of deviations.
Innovation Solution
Construct a time-series causal graph to depict the causality between predictive errors and their root causes, using a generative model to calculate the contribution of each variable to the predictive error, facilitating the interpretation of predictive models by analyzing the relationship between actual and predicted values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If predictive models using machine learning are adopted for infrastructure services, then prediction performance is improved, but interpretability and understanding of deviation causes deteriorate due to black-box nature
Solution Approach 1:
The patent introduces an intermediary explanation model that acts as a mediator between the black-box predictive model and users. This explanation model generates human-interpretable reasons for predictions, translating the opaque internal workings into understandable concepts without requiring changes to the original high-performance predictive model.
Solution Approach 2:
The patent segments the prediction explanation into multiple components: the main predictive model for accurate prediction, and separate explanation mechanisms that break down the reasoning into interpretable parts. This segmentation allows the system to maintain high prediction performance while providing transparent, segmentable explanations for each prediction.
2Loss of information
If existing explanation methods like SHAP and LIME are used, then some interpretability is provided, but causality among variables is not considered when determining deviation causes
Solution Approach 1:
The patent replaces the post-hoc explanation approach (mechanical addition of explanation layers) with a causality-aware generative model that inherently understands variable relationships. Instead of mechanically adding explanation after prediction, the system uses a generative model trained on causal relationships to produce explanations that reflect actual causality.
Solution Approach 2:
The patent changes the fundamental parameter of explanation generation from statistical correlation (as in SHAP/LIME) to causal relationship strength. By training the explanation model to recognize and prioritize causal relationships, the system transforms how explanations are generated, moving from mere association to genuine causality-based reasoning.
3Loss of information
If focus is placed on interpreting predictive model outputs, then model transparency is improved, but understanding of why predicted values deviate from actual values deteriorates
Solution Approach 1:
The patent introduces dynamic adaptivity in the explanation generation process. The explanation model dynamically adjusts its reasoning based on the specific prediction context, identifying which variables and relationships are most relevant for explaining deviations in each particular case, rather than providing static or generic explanations.
Solution Approach 2:
The patent implements a feedback mechanism where the explanation model continuously refines its causal reasoning based on the relationship between predicted and actual values. By analyzing deviation patterns and feeding this information back into the explanation generation, the system improves its understanding of deviation causes while maintaining model transparency.
Data Source
AI summary
An information processing apparatus includes processors. The processors construct a predictive model serving to receive respective pieces of first time-series data of input variables in a first period of time and predicts output variables to be obtained at a time point after the first period of time. The processors calculate third time-series data being time-series data of an index representing an error or a goodness of fit between (i) pieces of second time-series data representing the output variables predicted by the predictive model at time points included in the first period of time and (ii) correct time-series data representing correct answers to the output variables at the time points in the first period of time. The processors construct a time-series causal graph representing causality between the input variables and the index by using the pieces of first time-series data and the third time-series data.