Decision Tree Model Generation with Multi-Categorical Splits

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional decision tree technologies are inefficient in handling categorical variables and missing data, requiring cumbersome processes and resource-intensive methods, lacking a standard development process, and relying on non-standardized algorithms.

Innovation Solution

A computer-implemented method for generating machine learning decision trees that determines numeric variables, event rates, and multi-categorical splits for categorical variables, using training data to assign variables to nodes, and imputes missing values based on statistical significance, optimizing decision tree configuration and resource utilization.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If conventional decision tree technologies are used, then the model structure is simple, but the handling of categorical variables and missing data is ineffective

Engineering Contradiction:

Improvehandling of categorical variables and missing dataVSAvoidmodel development process complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent creates a universal decision tree development platform that handles multiple data types (categorical and numeric variables) and multiple scenarios (complete data and missing data) within a single integrated system. The platform provides standardized algorithms and procedures that work across different data characteristics, eliminating the need for separate handling methods for each variable type or data condition.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent transforms categorical variables into numeric representations through standardized encoding methods and adjusts data parameters through imputation techniques for missing values. By changing the representation parameters of categorical data and filling missing value parameters, the system converts diverse data formats into a unified structure that can be processed by standard decision tree algorithms.

Inventive Principle:

Principle #35Parameter changes

2Productivity

If ad-hoc approach with multiple algorithms is used, then flexibility is maintained, but resource utilization is inefficient

Engineering Contradiction:

Improvedecision tree development efficiencyVSAvoidresource consumption

Core Design Contradiction:

ProductivityVSQuantity of substance

Solution Approach 1:

The patent divides the decision tree development process into distinct, standardized segments: data preprocessing (handling categorical variables and missing data), model training, and evaluation. Each segment has its own optimized algorithm and procedures, allowing efficient processing of specific tasks without requiring multiple different algorithms throughout the entire development process.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The platform provides automated procedures that handle data preprocessing and model training without requiring manual intervention for each step. The standardized algorithms automatically detect data types, apply appropriate transformations, and configure model parameters, reducing the need for resource-intensive manual configuration and multiple algorithm trials.

Inventive Principle:

Principle #25Self-service

3Ease of operation

If multiple libraries with specific requirements are used, then algorithm flexibility is maintained, but the development process becomes cumbersome

Engineering Contradiction:

Improvedecision tree development processVSAvoidlibrary navigation complexity

Core Design Contradiction:

Ease of operationVSDevice complexity

Solution Approach 1:

The patent merges the functionality of multiple separate libraries into a single integrated platform. By combining data preprocessing tools, categorical variable handling methods, missing data imputation algorithms, and decision tree training functions into one unified system, the platform eliminates the need to navigate between multiple libraries with different requirements and interfaces.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS20240386287A1Methods and systems for developing decision tree machine learning models

Publication Date: 2024.11.21 CAPITAL ONE SERVICES LLC

AI summary

Systems, methods, articles of manufacture, and computer program products to generate decision trees are described. In some embodiments, a computer-implemented method to generate a machine learning decision tree model may include, via at least one processor of a computing device, determining a set of numeric variables for each of the plurality of categorical variables, determining an event rate for each of the plurality of categorical variables, determining, using training data, a plurality of splits for assigning the plurality of categorical variables to nodes of the machine learning decision tree model, wherein the plurality of splits comprises a plurality of multi-categorical splits assigning multiple of the plurality of categorical variables to a single node based on the event rate, and accessing data to generate the machine learning decision tree model comprising a plurality of nodes, at least a portion of the nodes assigned one of the plurality of multi-categorical splits.