Graph-Based Feature Subset Merging for ML Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning techniques face performance degradation and increased computational resources as data dimensionality increases, failing to accurately select relevant features and consider inter-dependencies, leading to sub-optimal model performance in applications like image recognition and fault detection.
Innovation Solution
A method that represents feature subsets as graphs, merges these graphs to form a merged feature graph, and selects a single subset of features based on edge weights and relationships, accounting for feature inter-dependencies to optimize model performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the number of features used in the model increases, then the model can capture more information and potentially improve accuracy, but the execution time and computational resources required increase significantly
Solution Approach 1:
The patent extracts and removes redundant or less important features from the dataset to create a reduced feature set. This is achieved through feature selection techniques that identify and eliminate features that do not contribute significantly to model performance, thereby reducing computational complexity and execution time while maintaining model accuracy.
Solution Approach 2:
The patent implements a hierarchical feature selection approach where features are organized in nested layers of importance. The system selects features in multiple stages, first identifying the most critical features, then progressively adding less important ones, allowing for optimized model training at different levels of feature granularity.
2Measurement precision
If the number of features used in the model increases, then more comprehensive data representation is achieved, but the computational resources required by the system increase
Solution Approach 1:
The system extracts only the most relevant features from the complete dataset based on their contribution to predictive power. By removing redundant features, the system reduces the computational burden and energy consumption associated with processing high-dimensional data while preserving the essential information needed for accurate modeling.
Solution Approach 2:
The patent applies partial action by selecting a subset of features that provides sufficient representation for the modeling task without processing all available features. This selective approach consumes fewer computational resources while achieving the necessary level of data representation quality for effective machine learning.
3Measurement precision
If the number of features increases, then the model can capture more patterns, but the behavior on real unseen data becomes less predictable
Solution Approach 1:
The system extracts features that demonstrate consistent predictive power across training and validation data, removing features that capture noise or overfit to training examples. This feature selection process improves generalization to unseen data by retaining only the most reliable pattern-capturing features.
Solution Approach 2:
The patent implements feedback mechanisms through cross-validation and performance monitoring on validation datasets. The system uses this feedback to iteratively refine feature selection, removing features that degrade generalization performance and retaining those that improve predictability on unseen data.
4Quantity of substance
If the number of features increases, then more information is available for modeling, but administrators have less knowledge about the most useful features in the system
Solution Approach 1:
The system extracts and highlights the most important features from the large feature set, presenting a manageable subset to administrators. This selective extraction maintains information availability for modeling while improving interpretability by focusing on the key features that drive model predictions.
Solution Approach 2:
The patent applies local quality by providing detailed information and interpretability for the most important features while summarizing or aggregating information about less important features. This approach allows administrators to understand the critical features in detail without being overwhelmed by the complete feature set.
5Quantity of substance
If the number of features increases, then the model can utilize more data dimensions, but the effects of feature inter-dependence in the system increase
Solution Approach 1:
The patent segments the feature selection process into multiple stages and groups features based on their inter-dependency relationships. By dividing the feature set into independent or weakly-dependent groups, the system reduces the complexity of managing inter-dependencies while preserving the beneficial effects of high-dimensional data representation.
Data Source
AI summary
A system, method and computer program product provides improved performance in machine learning, decision making and similar processes. In one example method, a plurality of individual subsets of features of a dataset comprising multiple features are received. The subsets may be provided by applying one or more feature selection methods to the dataset. Each subset is represented as a graph based on a predefined graph template. The example method merges the graphs of the plurality of individual subsets by overlaying the graphs on each other to form a merged feature graph. The merged feature graph may be used for identifying a single subset of features for use in machine learning, decision making and similar processes.