Rare Event Prediction via Bayes Decomposition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Predicting rare events in industrial applications is challenging due to the imbalance in data samples, as existing methods directly model the rare event, leading to inaccurate predictions.
Innovation Solution
The method decomposes the estimation problem into subproblems using Bayes' Theorem, considering the conjunction of multiple events, allowing for more accurate predictions by estimating the probability of a more common event and its conditional probability, which are then combined to approximate the original problem.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional methods directly model the rare event e, then the model structure is simple, but the prediction accuracy deteriorates due to data sample imbalance
Solution Approach 1:
The patent segments the rare event prediction problem into two separate subproblems: (1) predicting whether a common prerequisite event c occurs within time t, and (2) predicting whether the rare event e occurs given that c has occurred. This segmentation allows each subproblem to be solved with sufficient data samples, improving overall prediction accuracy despite increased model complexity.
Solution Approach 2:
The patent introduces a common prerequisite event c as an intermediary between the observation period and the rare event e. By modeling the conditional probability P(e|c) rather than directly modeling P(e), the system uses the intermediary event c to bridge the data gap, enabling accurate prediction of rare events through more abundant data on common events.
2Measurement precision
If the decomposition method is used to predict rare events, then prediction accuracy improves, but the computational complexity increases
Solution Approach 1:
The computational task is segmented into two independent model training processes: one for the common event c and one for the conditional event e given c. While this creates two models instead of one, each model can be trained efficiently on its respective data subset, and the overall computational burden is manageable compared to the alternative of requiring exponentially more data for direct rare event modeling.
3Adaptability or versatility
If hurdle models or two-part models are used, then the prediction framework is general, but they do not leverage the conjunction structure of events
Solution Approach 1:
The patent applies local quality by tailoring the modeling approach specifically to the conjunction structure of events. Instead of using a general two-part model that treats all zero-inflated data the same way, the system explicitly models the logical conjunction P(e) = P(e|c) × P(c) when events are conjunctions, thereby achieving both generality and precision for this specific event structure.
Data Source
AI summary
Example implementations described herein are directed to systems and methods for predicting if a conjunction of multiple events will occur within a certain time. It relies on an approximate decomposition into subproblems and a search among the possible decompositions and hyperparameters for the best model. When the conjunction is rare, the method mitigates the problem of data imbalance by estimating events that are less rare.