Unsupervised Decision Tree for Real-Time Attribute Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing decision tree algorithms are inefficient for predicting attribute values and validating input values due to their reliance on supervised learning, requirement of explicit target attributes, and the need to build multiple trees, leading to high computational costs and time consumption.
Innovation Solution
An unsupervised decision tree building algorithm that generates a single decision tree for records with multiple attributes, allowing for real-time prediction and validation without the need for a target attribute, using a combination of a single tree and indexing techniques for efficient data traversal.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised learning algorithms are used to build decision trees for prediction, then prediction accuracy can be improved, but computational cost and time consumption increase significantly
Solution Approach 1:
The patent segments the decision tree construction process by building separate trees for different target attributes independently, rather than using a single supervised learning model. This allows parallel processing and reduces overall computational time while maintaining prediction accuracy for each attribute.
Solution Approach 2:
The patent performs preliminary actions by pre-computing and storing intermediate results during the unsupervised tree construction phase. These pre-computed structures enable faster query processing and prediction without requiring re-computation during the prediction phase, thus reducing time consumption.
2Adaptability or versatility
If multiple decision trees are built for different target attributes, then prediction capability for multiple attributes is improved, but device complexity and computational resources increase
Solution Approach 1:
The patent creates a universal unsupervised decision tree structure that can serve multiple target attributes simultaneously. The same tree structure is reused for different prediction tasks by simply changing the target attribute specification, eliminating the need to build and store multiple separate trees for each attribute.
Solution Approach 2:
Instead of creating physically separate tree structures for each target attribute, the patent uses logical copying where the same tree structure is referenced multiple times with different target attributes. This reduces memory usage and structural complexity while maintaining the ability to predict multiple attributes.
3Measurement precision
If explicit target attributes are required for decision tree construction, then prediction focus is improved, but adaptability to unsupervised scenarios deteriorates
Solution Approach 1:
The patent inverts the traditional supervised learning approach by first building the decision tree structure in an unsupervised manner without specifying target attributes, and then assigning different attributes as targets during the query phase. This reversal enables the system to maintain prediction focus while achieving unsupervised learning capability.
Data Source
AI summary
A computer-implemented method to predict possible values for a subset of attributes of a record being interactively completed includes receiving a first value of a first attribute of the record. Further, from a decision tree, a first tree-level associated with the first attribute is determined. Further, in the decision tree, one or more nodes at a second tree-level are identified based on an index of the first tree-level. The index of the first tree-level includes a mapping between the first value and the nodes from the second tree-level based on historical records used to generate the decision tree data structure. Further, several paths in the decision tree are traversed, including a path from each of the nodes at the second tree-level towards a root node of the decision tree. The method also includes computing probabilities of the paths, and outputting values of the subset of attributes of the record along the path with highest probability as the possible values.