Hierarchical Supervised Dimensionality Reduction for Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing supervised learning methods fail to effectively utilize hierarchical relationships in data, leading to increased dimensionality and decreased model accuracy and interpretability when training models on flattened hierarchical data.
Innovation Solution
The approach identifies boundary equivalences between parent and child levels in hierarchical data, allowing for iterative selection of levels to reduce dimensionality, thereby training a decision tree model that substitutes or includes parent levels based on these equivalences, resulting in faster training and improved model accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If hierarchical data is flattened for model training, then data can be processed using conventional supervised learning methods, but dimensionality increases and model accuracy decreases
Solution Approach 1:
The patent segments the flattened hierarchical data by identifying and separating records that share common ancestor paths. This segmentation groups related records together, reducing the effective dimensionality while preserving hierarchical relationships. The segmentation allows the model to learn from structured patterns without being overwhelmed by the full flattened dimensionality.
Solution Approach 2:
The patent implements nesting by organizing data records according to their hierarchical ancestor paths, where child levels are nested within parent level contexts. This nested structure allows the model to capture hierarchical patterns efficiently, improving accuracy without requiring all flattened dimensions, thereby resolving the contradiction between ease of training and model accuracy.
2Ease of manufacture
If hierarchical data is flattened for model training, then data can be processed using conventional supervised learning methods, but training time increases
Solution Approach 1:
The patent extracts and removes redundant information from the flattened data by identifying records with common ancestor paths. By taking out only the necessary hierarchical information and discarding redundant repetitions, the data dimensionality is reduced, leading to faster training times while maintaining the ability to use conventional supervised learning methods.
Solution Approach 2:
The patent performs preliminary processing by pre-identifying and organizing records according to their ancestor paths before model training begins. This preliminary action structures the data in a way that reduces dimensionality upfront, so that subsequent training proceeds faster without sacrificing the ease of using standard supervised learning algorithms.
3Ease of manufacture
If random level selection is used in flattened data, then model training is simplified, but interpretability suffers
Solution Approach 1:
The patent applies universality by using ancestor path information that serves multiple functions: it provides structural organization for training, captures hierarchical relationships for accuracy, and offers interpretability through meaningful hierarchical labels. This multi-functional approach eliminates the need for random level selection, simultaneously achieving ease of training and interpretability.
Solution Approach 2:
The patent introduces ancestor paths as an intermediary structure that mediates between the flattened data format and the hierarchical relationships. This intermediary provides a bridge that maintains interpretability by preserving hierarchical context while allowing the use of conventional supervised learning methods on the flattened structure.
Data Source
Figure 1A~1B
Figure 2
Figure 3
AI summary
Systems and methods include identification first members of a child level of a dimension hierarchy which are associated with boundaries between second members of a parent level of the dimension hierarchy, training of a decision tree model based on data associated with the child level, extraction of predicates on the child level from the trained decision tree model, determination of a value based on the identified first members of the child level and on the extracted predicates on the child level, and determination, based on the value, whether to include the parent level and the child level within training data or to include the parent level and not include the child level within the training data.