Forecasting Engine With Directional Error Minimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing forecasting techniques fail to effectively minimize prediction errors in a preferred direction, leading to high deviations in forecasted values, particularly in applications where over-prediction is more dangerous than under-prediction, such as credit card or loan suitability screening.
Innovation Solution
A forecasting engine that learns a mapping between input attributes and target values using a custom error function to emphasize prediction errors in a preferred direction, iteratively constructing new models with reduced errors by weighting errors differently using a sigmoid function to minimize either positive or negative errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional regression analysis is used for forecasting, then the model can predict target values based on input attributes, but the prediction errors are not minimized in a preferred direction, leading to high deviation in forecasted values
Solution Approach 1:
The patent changes the error function parameters from symmetric (standard squared error) to asymmetric (custom error function with directionality). The custom error function assigns different weights to positive and negative errors based on business requirements, allowing the model to minimize errors in a preferred direction. This parameter transformation enables the model to achieve higher forecast accuracy by penalizing specific types of errors more heavily.
Solution Approach 2:
The patent introduces asymmetry in the error minimization process through the custom error function. Instead of treating all prediction errors equally (symmetric approach), the model applies asymmetric weighting where errors in the less desirable direction are penalized more heavily. This asymmetric error function allows the model to learn from historical data and adjust its predictions to minimize harmful errors while tolerating beneficial errors, thereby improving overall forecast reliability.
2Adaptability or versatility
If multiple regression models are used to reduce error, then the model can handle complex patterns, but the error present in the set of available signals is not reduced, resulting in high deviation in forecasted values
Solution Approach 1:
The patent makes the error function universal by designing a single custom error function that can handle multiple types of prediction scenarios. The error function incorporates a directionality parameter that can be adjusted based on business requirements, making it applicable to various forecasting problems without needing separate error functions for each scenario. This universal error function works with standard regression models while providing direction-specific error minimization capability.
3Device complexity
If a single model is used to reduce error, then the model structure remains simple, but the error minimized is to a lesser degree, resulting in insufficient accuracy improvement
Solution Approach 1:
The patent achieves enhanced error minimization effectiveness without increasing model structure complexity by changing the parameters of the error function. The custom error function introduces directionality parameters that can be adjusted to emphasize specific error types. This parameter-based approach allows the simple regression model structure to achieve superior prediction accuracy by modifying how errors are calculated and minimized, rather than adding complex model architecture.
Data Source
AI summary
The present invention provides a forecasting engine with the ability to minimize prediction error in a preferred direction. It comprises of a receiver configured to receive training data samples. In addition, the forecasting engine includes a building module configured to build a base learner model from the training data samples. In addition, the forecasting engine includes a custom error function that emphasizes prediction error along a pre-configured direction. In addition, the forecasting engine includes an error determination module configured to determine the prediction error made by the base learner model. In addition, the forecasting engine includes an error minimization module configured to construct a new model that has lesser prediction error than the base learner model, where prediction error is as defined by the custom error function. In addition, the forecasting engine includes an iteration module that manages multiple iterations of the error determination module and the error minimization module.