Business Prediction Model Using Rejected Sample Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional business prediction models based on machine learning exclude rejected business samples during training, leading to parameter estimation deviations and wasted labor costs, as they cannot predict future behaviors of rejected objects and determine their indicators.
Innovation Solution
A method that includes obtaining both passed and rejected business sample sets for training, using iterative analysis and screening to establish logistic regression models, and determining a final prediction model based on evaluation indicators to account for rejected samples, thereby reducing parameter estimation deviation and cost waste.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only business samples passing validation are selected for training, then the training process is simple and fast, but the model cannot predict future behaviors of rejected objects and suffers from parameter estimation deviation
Solution Approach 1:
The training process is segmented into multiple stages: initial model training using only passed samples, then iterative refinement by incorporating rejected samples that meet specific criteria. This segmentation allows the model to first establish a baseline and then progressively improve accuracy by selectively adding complex rejected sample data.
Solution Approach 2:
Instead of using all rejected samples (excessive action), the method selectively incorporates only those rejected samples that meet specific criteria such as falling within certain probability score ranges or satisfying business rules. This partial action approach improves parameter estimation accuracy without overwhelming the training process with irrelevant data.
2Productivity
If rejected business samples are excluded from training, then the training data is clean and manageable, but labor costs are wasted and prediction accuracy for rejected objects is lost
Solution Approach 1:
The method extracts valuable information from rejected samples by identifying and separating those that contain useful predictive patterns. Specifically, rejected samples are analyzed and those meeting certain criteria (e.g., probability scores within specific ranges, satisfaction of business rules) are extracted and incorporated into the training set, while others are discarded.
Solution Approach 2:
Business rules and probability score thresholds act as intermediaries to mediate between passed and rejected samples. These intermediaries determine which rejected samples should be incorporated into training by evaluating them against predefined criteria, thus bridging the gap between clean training data and comprehensive prediction capability.
3Measurement precision
If all rejected samples are incorporated into training, then parameter estimation accuracy improves, but the training process becomes overly complex and time-consuming
Solution Approach 1:
The method applies partial action by incorporating only a subset of rejected samples that meet specific criteria rather than all rejected samples. This selective approach includes samples with probability scores within certain ranges or those satisfying business rules, thereby improving prediction accuracy without proportionally increasing training time.
Solution Approach 2:
The method changes parameters by introducing probability score thresholds and business rule criteria as filtering mechanisms. By adjusting these parameters (thresholds, ranges, rules), the training process can dynamically control which rejected samples are incorporated, optimizing the balance between prediction accuracy and training efficiency.
Data Source
AI summary
A business prediction method includes: obtaining a first business sample set and a second business sample set; performing training based on the first business sample set and the second business sample set to obtain a business prediction model, and predicting received to-be-predicted business information based on the business prediction model to obtain a business prediction result corresponding to the received to-be-predicted business information. A business prediction apparatus is further provided. The business prediction method and the business prediction apparatus take into account data features of some business samples of being rejected in a business validation, while considering business samples of passing the business validation. This restores a business scenario, reduces the waste of costs of the rejected samples, and balances demands for a modeling sample and a rejected sample reasonably when there are insufficient samples of passing the business validation.