Automated Feature Selection for Linear Models

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current data analytics applications require extensive training in data science for users to select features effectively for model generation, leading to high processing and memory overhead, especially with large datasets, resulting in significant latency and limited interactivity.

Innovation Solution

Implementing techniques such as automatic feature selection using L1 regularizations and the LASSO regression procedure, along with a rake optimization technique, to determine a close-to-optimal linear model without storing the dataset in memory, allowing for fast and interactive analysis of large datasets.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If users manually select features for model generation, then model accuracy may be improved, but user expertise requirements and time consumption increase significantly

Engineering Contradiction:
Improvemodel accuracyVSAvoiduser expertise requirements
Core Design Contradiction:
Measurement precisionVSEase of operation

Solution Approach 1:

The system performs automatic feature selection using L1 regularizations and LASSO regression procedures, allowing the data itself to determine the most relevant features without requiring user expertise in data science. The algorithm automatically identifies and selects features that contribute most to model accuracy, making the system self-sufficient in the feature selection process.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system changes the parameter of feature selection from manual user choice to automated algorithmic determination. By implementing L1 regularizations and LASSO regression, the system transforms the feature selection process into a computational optimization problem that can be solved automatically, reducing dependency on user expertise while maintaining or improving model accuracy.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If complete datasets are stored in memory for analysis, then processing accuracy is maintained, but memory overhead and system resource consumption increase significantly

Engineering Contradiction:
Improveprocessing accuracyVSAvoidmemory overhead
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The system extracts only the essential information needed for model generation from the complete dataset, rather than storing and processing the entire dataset in memory. By using LASSO regression and automatic feature selection, the system identifies and extracts only the most relevant features and corresponding data points, significantly reducing memory overhead while maintaining processing accuracy for the selected features.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The system segments the large dataset into manageable portions by focusing only on the subset of data corresponding to the selected features. Instead of loading the complete dataset into memory, the system processes segmented portions of data that are relevant to the model generation task, reducing overall memory consumption while maintaining analytical accuracy.

Inventive Principle:
Principle #1Segmentation

3Adaptability or versatility

If extensive feature lists are provided for user selection, then comprehensive analysis coverage is achieved, but processing time and latency increase significantly

Engineering Contradiction:
Improveanalysis coverageVSAvoidprocessing time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system performs preliminary feature selection using L1 regularizations and LASSO regression before the main model generation process. This preliminary action automatically identifies and ranks the most relevant features from the complete feature list, reducing the number of features that need to be processed in subsequent steps. This pre-processing step maintains comprehensive analysis coverage by considering all features initially, but reduces processing time by focusing computational resources on the most important features.

Inventive Principle:
Principle #10Preliminary action

4Ease of operation

If interactive data analytics is implemented for large datasets, then user flexibility is improved, but processing latency and system responsiveness deteriorate

Engineering Contradiction:
Improveuser flexibilityVSAvoidprocessing speed
Core Design Contradiction:
Ease of operationVSSpeed

Solution Approach 1:

The system provides interactive data analytics capabilities through automated feature selection and model generation, allowing users to flexibly analyze different datasets without requiring deep data science expertise. The automated algorithms handle the computationally intensive tasks of feature selection and model training, enabling users to interact with the system flexibly while the system maintains fast processing speeds through efficient algorithmic implementation.

Inventive Principle:
Principle #25Self-service

Data Source

PatentUS11699094B2Automatic feature selection and model generation for linear models
Publication Date: 2023.07.11 SALESFORCE INC
  • US11699094B2 patent drawing
  • US11699094B2 patent drawing
  • US11699094B2 patent drawing

AI summary

Methods, systems, and devices for automated feature selection and model generation are described. A device (e.g., a server, user device, database, etc.) may perform model generation for an underlying dataset and a specified outcome variable. The device may determine relevance measurements (e.g., stump R-squared values) for a set of identified features of the dataset and can reduce the set of features based on these relevance measurements (e.g., according to a double-box procedure). Using this reduced set of features, the device may perform a least absolute shrinkage and selection operator (LASSO) regression procedure to sort the features. The device may then determine a set of nested linear models—where each successive model of the set includes an additional feature of the sorted features—and may select a “best” linear model for model generation based on this set of models and a model quality criterion (e.g., an Akaike information criterion (AIC)).