Segmented ML Modeling for Predictive Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning-based models often underfit or overfit predictive features in data with interdependencies, leading to inadequate trend-based analytics for forecasting future outcomes, and are prone to errors from anomalous data.
Innovation Solution
The system trains multiple machine learning models, including a base model and segmented models focused on specific subsets of features, and a merged model that combines outputs to capture interdependencies and mitigate anomalous data through short-term and long-term attrition definitions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single machine learning model is trained on all features, then the model complexity is low, but the model underfits predictive features due to interdependencies in the data
Solution Approach 1:
The patent divides the single machine learning model into multiple segmented models, where each model is trained on a specific subset of features related to particular segments (e.g., device types, locations, sectors). This segmentation allows each model to specialize in capturing interdependencies within its specific feature subset, thereby improving predictive accuracy while maintaining manageable model complexity through modular design.
2Measurement precision
If multiple segmented models are trained on feature subsets, then predictive accuracy improves by capturing interdependencies, but the system complexity increases
Solution Approach 1:
The patent combines multiple segmented models into a unified ensemble system that integrates their predictions. The merging mechanism aggregates outputs from individual segmented models, allowing the system to leverage the specialized predictive capabilities of each model while presenting a cohesive solution. This approach maintains high predictive accuracy by capturing interdependencies through the ensemble while managing system complexity through structured integration.
Solution Approach 2:
The ensemble of segmented models serves multiple functions: each segmented model handles specific feature subsets and segments, while the combined system provides comprehensive predictions across all entities. This multi-functionality allows the system to adapt to different entity types, feature combinations, and prediction scenarios, improving predictive accuracy without proportionally increasing complexity.
3Reliability
If smoothing or normalization is applied to mitigate anomalous data, then the impact of anomalies is reduced, but too much smoothing leads to loss of important predictive signals
Solution Approach 1:
The patent applies different processing treatments to different segments and features based on their specific characteristics. Rather than uniformly smoothing all data, the system identifies and handles anomalous data locally within specific segments or feature subsets, preserving important predictive signals in other areas. This localized approach maintains reliability by addressing anomalies where they occur without unnecessarily smoothing away valuable information elsewhere in the dataset.
Data Source
AI summary
The disclosure relates to systems and methods of generating behavior classifications that predict a behavior of an entity by training and executing a base machine learning (ML) model, a plurality of segmented ML models, and a merged ML model. Training data may be historical entity data, which may be grouped into different segments that describe the entity. The base ML model may be trained to predict entity behavior across a plurality of segments. Each segmented ML model may be trained to the generate a segmented behavior class that predicts entity behavior based on a respective segment. A system may provide the base class and the plurality of segmented classes as input to a merged model that was trained based on weights for each of the base ML model and the plurality of segmented ML models to generate a behavior classification representing a prediction of the entity behavior.