Feature Graph Visualization for Automated Feature Selection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Automated feature engineering generates numerous features, making it challenging to efficiently select and interpret which features are useful for machine learning algorithms, leading to issues like overfitting, data leakage, and systematic bias.

Innovation Solution

Converting the output of automated feature engineering into interpretable Boolean expressions and visualizing them as a connected feature graph to assess feature utility and potential bias, using techniques like Boolean simplification, information gain measurement, and similarity metrics to rank and prune features.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If automated feature engineering is applied to generate numerous features, then the quantity of candidate features increases, but the interpretability and selection efficiency deteriorate

Engineering Contradiction:

Improvequantity of candidate featuresVSAvoidfeature selection efficiency

Core Design Contradiction:

Quantity of substanceVSEase of operation

Solution Approach 1:

The patent segments the feature selection process into distinct phases: automated feature engineering generates candidate features, then feature selection routines evaluate and filter them. The system divides features into groups based on their properties and evaluates them separately, making the overall selection process more manageable and efficient despite the large quantity of candidate features.

Inventive Principle:

Principle #1Segmentation

2Manufacturing precision

If feature selection routines are applied to evaluate generated features, then the quality of feature selection improves, but the computational time and complexity increase

Engineering Contradiction:

Improvefeature selection qualityVSAvoidfeature selection time

Core Design Contradiction:

Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent applies preliminary filtering and evaluation criteria before conducting comprehensive feature selection. The system pre-processes candidate features by applying basic quality checks and filtering out obviously poor features before subjecting them to more computationally intensive selection routines, thereby reducing overall computation time while maintaining selection quality.

Inventive Principle:

Principle #10Preliminary action

3Reliability

If complex automated feature engineering is used to generate millions of features, then the predictive power increases, but the interpretability and bias detection capability decrease

Engineering Contradiction:

Improvepredictive powerVSAvoidinterpretability information

Core Design Contradiction:

ReliabilityVSLoss of information

Solution Approach 1:

The patent introduces feature graphs as an intermediary representation that connects complex generated features to their constituent raw features and transformations. These graphs serve as a mediator that preserves interpretability information by visually representing the provenance and relationships of features, allowing users to trace back complex features to their original data sources and understand how they were constructed.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS20230267107A1Graphical representation of automated feature engineering for feature selection

Publication Date: 2023.08.24 MANGARELLA MICHAEL CHARLES

AI summary

Systems and methods are provided that convert the output of automated feature engineering techniques into interpretable Boolean expressions that can be visualized as a connected feature graph.