Decision Tree Leaf Accuracy via Random Parameter Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine-learning models, such as decision tree models used in search engines, face challenges in comprehension and overfitting, leading to suboptimal accuracy and resource inefficiency in processing power and energy usage.
Innovation Solution
A method is introduced to determine the accuracy parameter of a trained decision tree prediction model by generating a subset of random parameters and associating them with leaves in the model, allowing for improved leaf accuracy parameter determination and overall model accuracy without exceeding acceptable CPU processing thresholds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional machine-learning models (decision tree models) are used in search engines, then the models can process search queries and provide predictions, but the models suffer from comprehension difficulties and overfitting leading to suboptimal accuracy
Solution Approach 1:
The patent changes the parameter representation by generating random parameters associated with leaves in the decision tree model. Instead of using fixed or deterministic parameters, the system introduces stochastic elements by generating random parameters within specific ranges, which helps prevent overfitting while maintaining prediction accuracy. This parameter transformation allows the model to generalize better across different search queries.
2Measurement precision
If more computational resources are allocated to improve model accuracy, then prediction accuracy improves, but CPU processing power and energy usage increase beyond acceptable thresholds
Solution Approach 1:
The patent applies partial action by generating random parameters only for specific leaves in the decision tree rather than for the entire model. This selective approach allows the system to improve accuracy where most needed while avoiding unnecessary computational overhead. The random parameters are generated within controlled ranges and applied selectively, balancing accuracy improvement with resource conservation.
Solution Approach 2:
By transforming parameters to random values within specific ranges, the system achieves better generalization without requiring increased computational resources for model training or inference. This parameter transformation approach improves prediction accuracy while maintaining efficient CPU usage, as the random parameter generation is computationally lightweight compared to retraining the entire model.
3Measurement precision
If the model uses more factors and features to improve prediction accuracy, then the model becomes more comprehensive, but the device complexity and resource requirements increase
Solution Approach 1:
The patent simplifies model complexity by using random parameter generation instead of introducing additional factors or features. Rather than expanding the model structure with more nodes, branches, or input features, the system transforms existing leaf parameters into random values within specified ranges. This approach maintains model structural simplicity while improving prediction accuracy through stochastic parameter transformation.
Data Source
AI summary
A computer-implemented method of and a system for generating a prediction model and determining an accuracy parameter of a trained decision tree prediction model. The method comprises accessing the trained decision tree prediction model having been generated at least partially based on a set of training objects; generating a subset of random parameters of interest; associating the subset of random parameters of interests with a given leaf; determining a leaf accuracy parameter based on (i) the parameters of interest associated with the given leaf and (ii) the subset of random parameters of interest of the given leaf; and determining the accuracy parameter of the trained decision tree prediction model based on the determined leaf accuracy parameter for each of the leafs of the decision tree.