Prediction Visualization for Time-Series Component Attribution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to efficiently enhance user understanding of the logic underlying predicted values in time-series data visualizations, leading to a lack of trust in the predictive algorithms.
Innovation Solution
A visualization system that decomposes time-series data into multiple components, including trend, cycle, influencer, and fluctuation components, and presents their contributions to the predicted value using a waterfall chart, providing a clear local and global explanation of the prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If global characterizations of the algorithm are provided to enhance understanding, then user comprehension improves, but the logic underlying predicted values remains obscure and difficult to understand
Solution Approach 1:
The patent segments the predicted value into multiple additive components (trend component, cyclic component, seasonal component, residual component, and influencer components). Each component is visualized separately in the chart, allowing users to understand the contribution of each factor to the overall prediction without being overwhelmed by the complex algorithm logic.
Solution Approach 2:
The patent introduces an intermediary visualization layer (the chart showing component contributions) that mediates between the complex algorithm and the user. This intermediary translates the algorithm's internal logic into an easily interpretable visual format, making the underlying logic transparent without requiring users to understand the algorithm itself.
2Reliability
If detailed component breakdowns are provided for each predicted value, then user trust increases, but visualization complexity increases
Solution Approach 1:
The patent merges multiple pieces of information (predicted value, component contributions, component types, and visual representations) into a single unified chart. This consolidation provides comprehensive information for building user trust while maintaining visual simplicity through an integrated rather than fragmented presentation.
Solution Approach 2:
The patent applies local quality by providing detailed component breakdowns specifically for each predicted value location in the time series, while maintaining overall visualization coherence. Each point in the chart can be independently examined for detailed component contributions without complicating the entire visualization structure.
Data Source
AI summary
A system and method include reception of a request for a predicted value of a target over a time period, determination of first time-series data associated with the target, determination of a plurality of time-series components of the first data, determination of the predicted value of the target over the time period based on the plurality of time-series components, generation of a visualization including the predicted value and a contribution of each of the time-series components to the predicted value, and transmission of the visualization to a remote device.