Decision Tree Pruning for Academic Ability Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional academic ability estimation is complicated and inaccurate due to the need for comprehensive learning data, which becomes obsolete over time, leading to degradation in estimation accuracy.
Innovation Solution
An academic ability estimation model generation device that includes a decision tree generation unit, pruning unit, and category generation unit, which generates a decision tree using correct/incorrect-answer information and prunes leaf nodes with low entropy, setting new terminal ends as categories, and connects auxiliary decision trees to improve estimation accuracy without requiring comprehensive learning data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If comprehensive learning data is collected for academic ability estimation, then the coverage of estimation is improved, but the complexity of the system increases and data becomes obsolete over time
Solution Approach 1:
The invention extracts only the necessary features from comprehensive learning data by using decision tree analysis to identify key questions that discriminate between different academic ability levels. Instead of collecting and processing all learning data, the system extracts a minimal set of critical questions that provide sufficient discrimination power for accurate estimation.
Solution Approach 2:
The invention creates a simplified model (decision tree structure) that copies only the essential discriminatory patterns from the comprehensive learning data. This model captures the key relationships between questions and ability levels without requiring the original comprehensive data set, enabling accurate estimation with minimal data collection.
2Ease of manufacture
If past learning data is used for estimation, then data collection is simplified, but the accuracy degrades over time due to obsolescence
Solution Approach 1:
The invention performs preliminary analysis to identify which specific questions have the highest discriminatory power for academic ability estimation. By pre-selecting these key questions and building a decision tree model based on them, the system prepares a streamlined estimation framework that maintains accuracy while simplifying ongoing data collection to only the essential questions.
3Measurement precision
If detailed classification categories are created, then the precision of ability level identification is improved, but the complexity of category management increases
Solution Approach 1:
The invention segments the continuous spectrum of academic ability into discrete classification categories based on decision tree leaf nodes. Each leaf node represents a distinct ability level category defined by specific patterns of question answers. This segmentation provides precise classification while managing complexity through the hierarchical structure of the decision tree, which automatically organizes categories based on discriminatory power.
Data Source
AI summary
An academic ability estimation model generation device to generate an academic ability estimation model with which current academic ability is accurately estimated without requiring comprehensive learning data. The academic ability estimation model generation device includes a decision tree generator that generates a decision tree by using correct/incorrect-answer information as teacher data, the correct/incorrect-answer information indicating that a plurality of answerers who have answered a question group consisting of a plurality of predetermined questions have answered each question correctly or incorrectly; a pruner that deletes a leaf node when an entropy of a classification result indicated by the leaf node being a terminal end of the decision tree which is generated is equal to or lower than a predetermined value. Further, there is a category generator that sets each new terminal end of the decision tree after deleting the leaf node as a category to which any of the answerers belongs.