Forecasting Model Granularity Mismatch via Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Forecasting models face accuracy issues when predicting at finer levels of granularity due to sparse data, often resulting in predictions being made at coarser levels of granularity or with deemphasized accuracy.
Innovation Solution
The approach involves receiving a target level of granularity, performing forecast modeling at an aggregated level, and determining a distribution scheme to distribute the results to the target level, using methods such as aggregation schemes like sum, average, most frequent value, or concatenation of unique values, and hyperparameter tuning to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If forecasting is performed at finer levels of granularity, then prediction detail and usefulness are improved, but accuracy deteriorates due to sparse data
Solution Approach 1:
The patent segments the forecasting process into two distinct stages: (1) aggregation stage where data from multiple fine-grained sources are combined to form aggregated time series data, and (2) distribution stage where the aggregated forecast results are allocated back to individual fine-grained targets. This segmentation allows the model to operate at an aggregated level where data is sufficient while still producing fine-grained forecasts.
Solution Approach 2:
The patent introduces aggregated time series data as an intermediary between the sparse fine-grained input data and the desired fine-grained forecast outputs. By first creating aggregated forecasts from sufficient data and then distributing these to individual targets, the system mediates the data scarcity problem while maintaining fine-grained prediction capability.
2Quantity of substance
If forecasting is performed at aggregated levels, then data sufficiency is improved, but prediction detail and usefulness deteriorate
Solution Approach 1:
The patent changes the granularity parameter dynamically during the forecasting process. The model trains and performs aggregation at a coarser granularity level where data is abundant, then applies distribution algorithms to transform the results to the desired fine granularity level. This parameter transformation allows the system to leverage data at one level while delivering results at another level.
3Adaptability or versatility
If distribution methods are added to extend forecasting to finer granularity, then system complexity increases, but forecasting capability at fine granularity is improved
Solution Approach 1:
The patent creates a universal forecasting framework that can handle multiple granularity levels through a single aggregated forecasting model. The distribution component serves multiple functions: it can allocate forecasts to different fine-grained targets, handle various distribution schemes (proportional, equal, etc.), and adapt to different data types. This multi-functionality reduces the need for separate models for each granularity level.
Data Source
AI summary
Aspects of the disclosure are directed to an approach for extending forecasting models to various levels of granularity. The approach can include receiving a target level of granularity for distributing a forecast, performing forecast modeling at an aggregated level of granularity, and determining a distribution method to distribute results of the forecast model at the target level of granularity. The approach can improve performance over existing forecasting models with minimal overhead.