Automated Data Hierarchy Extraction Through Low-Count Class Pruning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional machine learning classifiers trained on data sets with classes based on few observations produce inaccurate or uninformative classifications due to low signal-to-noise ratios, leading to misleading data analysis and reduced predictive ability.
Innovation Solution
A system that generates a hierarchical classification by pruning categories with insufficient data items, reassigning them to higher-level categories, and training a machine learning model using the revised data set to improve classification accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional classifiers are trained on data sets with classes based on few observations, then the classification process can be performed, but the classification accuracy and predictive ability deteriorate due to low signal-to-noise ratios
Solution Approach 1:
The patent segments the classification hierarchy by identifying and separating classes with insufficient observations from the main classification structure. This is achieved through detecting classes that fall below a minimum observation threshold and creating a segmented training set that excludes these problematic classes, thereby improving overall classification accuracy while maintaining the structure of well-supported classes
Solution Approach 2:
The patent applies preliminary action by performing class observation analysis and filtering before the actual machine learning training process. The system pre-identifies classes with insufficient observations and removes them from the training data set beforehand, preventing these low-signal classes from degrading the model's predictive ability during training
2Adaptability or versatility
If classes with very few observations are included in the training data set, then the classifier can recognize more categories, but the model produces inaccurate or uninformative classifications due to statistically questionable data
Solution Approach 1:
The patent changes the parameter of class inclusion criteria by introducing a minimum observation threshold. This parameter modification filters out classes that do not meet the statistical reliability requirement, ensuring that only classes with sufficient observations are included in the training set, thereby maintaining both versatility and reliability
Solution Approach 2:
The patent implements feedback through an iterative process where the system evaluates class observation counts, identifies problematic classes, removes them from the training set, retrains the model, and evaluates performance. This feedback loop continues until the training data set contains only classes that meet the minimum observation threshold, ensuring reliable classifications
3Measurement precision
If manual classification is performed on high volume data, then classification accuracy can be maintained, but the processing time and labor requirements become prohibitive
Solution Approach 1:
The patent applies preliminary action by pre-processing the training data to identify and remove classes with insufficient observations before model training. This preliminary filtering step creates a cleaner, more reliable training set that enables automated machine learning models to achieve accuracy comparable to manual classification while maintaining high processing throughput
Solution Approach 2:
The patent substitutes the mechanical manual classification process with an automated machine learning system. By replacing human analysts with an automated model trained on filtered data, the system achieves both high accuracy and high productivity, processing millions or billions of data items per hour without the labor constraints of manual classification
Data Source
AI summary
Techniques are disclosed for revising training data used for training a machine learning model to exclude categories that are associated with an insufficient number of data items in the training data set. The system then merges any data items associated with a removed category into a parent category in a hierarchy of classifications. The revised training data set, which includes the recategorized data items and lacks the removed categories, is then used to train a machine learning model in a way that avoids recognizing the removed categories.