Time Series Forecast Explanation via Data Point Perturbation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for explaining time series predictions made using statistical models, such as Exponential Smoothing and ARIMA models, are limited and require manual, computationally intensive processes, lacking efficient techniques for providing insight into the importance of individual data points on forecasts.
Innovation Solution
A technique is developed to generate explanation information for time series forecasts by perturbing data points in the training dataset, using the trained model to compute feature forecast weights and importance values, which are then visualized to indicate the impact of each data point on the forecast, applicable to various linear statistical models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If existing explanation methods (LIME, DeepLIFT, etc.) are used for time series forecasts, then interpretability is improved, but applicability is limited to certain model types only
Solution Approach 1:
The patent creates a universal explanation framework that works across multiple statistical model types (ARIMA, Exponential Smoothing, regression models) through a common perturbation-based interface. The system defines standardized explanation outputs (feature weights, importance scores) that can be generated for any model implementing the perturbation interface, making the explanation capability universally applicable rather than model-specific.
2Measurement precision
If manual explanation generation is performed for statistical models, then accuracy of explanation is improved, but productivity deteriorates due to computational intensity
Solution Approach 1:
The patent implements efficient explanation generation by computing perturbations selectively rather than exhaustively. The system uses sampling strategies where only a subset of data points are perturbed, and perturbation magnitudes are optimized to achieve sufficient explanation quality without requiring complete enumeration of all possible perturbations, thus balancing accuracy with computational efficiency.
3Loss of information
If comprehensive explanation information is provided for all data points, then interpretability is improved, but device complexity increases
Solution Approach 1:
The patent extracts and prioritizes only the most important features for explanation rather than providing equally detailed information for all data points. The system computes importance scores for each feature and selectively highlights top-contributing features, removing unnecessary complexity from less important elements while maintaining interpretability for the critical factors driving the forecast.
Data Source
AI summary
Techniques are described for providing explanation information for time series-based predictions made using statistical models, such as linear statistical models, examples of which include various Exponential Smoothing models, Autoregressive Integrated Moving Average (ARIMA) models, and others. For a forecast predicted by a statistical model that has been trained upon and/or fit to a set of historical times series data points, an explanation is generated for the forecast, where the explanation for the forecast includes information indicative of the importance or impact or influence of individual time series data points in the set on the forecast. The explanation for the forecast may be output to a user along with the forecast. This enables the user to have some visibility into why the particular forecast was predicted by the statistical model.