Stage-Specific Processing Pipeline Models for Propensity Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems fail to accurately predict a user's propensity to obtain an item or service due to their aggregate approach, which does not account for individual stages in the processing pipeline, leading to inefficient computing and network resource usage and inaccurate corrective actions.
Innovation Solution
A machine learning model that predicts user propensity and performs corrective actions at individual stages of the processing pipeline, using feature selection and gradient boosting algorithms like XGBoost, with shapley values and Bayesian optimization to enhance accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If an aggregate approach is used to predict user propensity, then computing and network resources are consumed, but prediction accuracy deteriorates due to not accounting for individual stages in the processing pipeline
Solution Approach 1:
The patent segments the processing pipeline into multiple discrete stages (e.g., initial application, underwriting, approval). Instead of treating all users uniformly, the system identifies which stage each user is in and applies stage-specific prediction models. This segmentation allows the system to capture nuanced variations in user behavior at different pipeline points, thereby improving prediction accuracy without requiring a complete redesign of the entire system.
Solution Approach 2:
The patent implements local quality by training separate machine learning models for each stage of the processing pipeline. Each model is specialized to predict propensity scores for users at that particular stage, using features and patterns relevant to that specific context. This localized approach ensures that predictions are tailored to the specific characteristics of each pipeline stage, improving overall accuracy while maintaining manageable model complexity.
2Measurement precision
If individual user characteristics and dynamic conditions are considered at each pipeline stage, then prediction accuracy improves, but computational resource consumption increases
Solution Approach 1:
The patent applies preliminary action by pre-training multiple stage-specific models offline using historical data. During runtime, the system only needs to infer which stage a user is in and apply the corresponding pre-trained model, rather than performing comprehensive analysis from scratch. This preliminary preparation significantly reduces real-time computational resource consumption while maintaining high prediction accuracy.
Solution Approach 2:
The patent changes parameters by using different feature sets and model configurations optimized for each pipeline stage. Instead of using a single monolithic model with all possible features, the system selectively activates appropriate features and models based on the user's current stage. This parameter adaptation reduces computational overhead by processing only relevant data for each specific prediction context.
3Reliability
If corrective actions are delayed until the end of the processing pipeline, then system complexity is reduced, but the ability to mitigate risks from changing conditions deteriorates
Solution Approach 1:
The patent implements preliminary action by enabling corrective actions at any stage of the processing pipeline based on real-time propensity predictions. When the system predicts a user is likely to obtain an item despite changing conditions (such as interest rate changes), it can immediately trigger appropriate corrective actions (e.g., notifying underwriters, adjusting terms) rather than waiting until the end. This stage-aware approach significantly improves risk mitigation capability by enabling timely interventions.
Solution Approach 2:
The patent incorporates feedback mechanisms where prediction results at each stage feed into decision-making processes. The system continuously monitors propensity scores and uses this feedback to trigger corrective actions when needed. This feedback loop allows the system to adapt to changing conditions dynamically, improving reliability by ensuring risks are addressed promptly rather than discovered too late in the pipeline.
Data Source
AI summary
An example method includes receiving, via a network interface, data relating to a user and a processing pipeline relating to obtaining a first item; determining a current state of the user in the processing pipeline; inputting the received data and the current state into a machine learning model that is trained to receive such inputs for a particular user and generate an output specifying a propensity that the particular user will obtain a particular item; in response to inputting the received data and current state, obtaining, from the machine learning model, a model output specifying a propensity that the user will obtain the first item; and performing, based on the propensity that the user will obtain the first item, a corrective action that mitigates for risks of changing conditions and corresponding impact on an electronic platform when the user obtains the first item.