Multimodel Forecasting Weighting via Information Criteria
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deciding which statistical forecasting models or their combinations to use for predicting future values is challenging, especially when multiple models are available for each item, and existing methods lack an efficient way to generate weights for combining model outputs effectively.
Innovation Solution
A system and method for automatically generating a weighted average forecast model by optimizing multiple forecasting models using time series data and calculating weighting factors based on information criteria indicative of fit quality, allowing for improved predictive performance by combining model outputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple forecasting models are combined to improve predictive performance, then forecasting accuracy is improved, but model selection complexity increases
Solution Approach 1:
The patent transforms the complex model selection problem into a parameter optimization problem by using information criteria values (such as AIC, BIC) to quantify model fit quality. By converting qualitative model performance assessment into quantitative parameter comparison, the system automatically determines optimal model combinations and weighting factors, thereby improving forecasting accuracy while managing selection complexity through mathematical parameterization.
Solution Approach 2:
The system implements self-service by enabling automated model selection and weight assignment without requiring manual expert intervention. The information criteria-based framework automatically evaluates multiple forecasting models, computes their relative performance metrics, and generates optimal combining weights, allowing the system to serve itself in the model selection process and reducing human involvement in complex model selection tasks.
2Productivity
If automated algorithms are used to generate predictions for large numbers of items, then productivity is improved, but the challenge of selecting appropriate models for each item increases
Solution Approach 1:
The patent applies universality by creating a universal model selection framework that can handle diverse forecasting models (ARIMA, exponential smoothing, regression models, etc.) through a common information criteria-based evaluation mechanism. This multi-functional approach allows the same automated algorithm to selectively apply appropriate models across thousands of different items, maintaining high productivity while managing model selection complexity through a unified evaluation standard.
Solution Approach 2:
The system uses parameter changes by transforming the model selection challenge into a parameter optimization task. By computing information criteria values as quantitative parameters for each model-item combination, the system enables automated algorithms to efficiently compare and select models based on numerical parameters rather than qualitative assessment, thereby scaling productivity across large numbers of items while controlling selection complexity through mathematical parameterization.
Data Source
AI summary
Systems and methods are provided for a computer-implemented method for automatically generating a weighted average forecast model that includes receiving a plurality of forecasting models and time series data. At least one parameter of each of the received forecasting models is optimized utilizing the received time series data. A weighting factor is generated for each of the plurality of optimized forecasting models utilizing an information criteria value indicating fit quality of each of the optimized forecasting models, and the generated weighting factors are stored.