Naive Bayes Classifier Training Set Balancing via LCM Replication
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Naïve Bayes classifiers are sensitive to imbalances in training statement distributions, leading to skewed classification and reduced accuracy, as they do not account for the varying number of training statements required for each class, resulting in potential misclassifications.
Innovation Solution
The method calculates the lowest common multiple of training statements for each class and replicates training statements to balance the distribution, ensuring higher confidence levels in classification and flattening certainty distributions for untrained inputs without modifying the algorithm or reducing training statements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If training statements are used in their original distribution, then the classifier can process real-world data efficiently, but the classification accuracy deteriorates due to class imbalance
Solution Approach 1:
The patent creates copies of training statements from underrepresented classes to balance the training set. Specifically, it identifies the lowest common multiple of training statement counts across classes and replicates statements from classes with fewer samples until the distribution is balanced, thereby improving classification accuracy without modifying the algorithm itself
Solution Approach 2:
The patent changes the distribution parameter of the training set by balancing the number of training statements across different classes. It calculates the lowest common multiple of training statement counts and adjusts the distribution to ensure each class has an equal number of samples, which resolves the class imbalance issue while maintaining the original algorithm
2Measurement precision
If training statements are replicated to balance distribution, then predictive accuracy improves, but the complexity of data preparation increases
Solution Approach 1:
The patent implements a self-service mechanism where the system automatically calculates the lowest common multiple of training statement counts, identifies which classes need replication, and performs the balancing process without manual intervention. This automated approach reduces the perceived complexity for users while achieving balanced training sets
Solution Approach 2:
The patent performs data balancing as a preliminary step before actual model training. By pre-calculating the lowest common multiple and replicating training statements in advance, it prepares balanced training sets that can be directly used for training, thereby improving predictive accuracy while containing the complexity within a one-time preprocessing operation
Data Source
AI summary
In an approach to improving the predictive accuracy of classifiers, one or more computer processors calculate one or more training set statistics. The one or more computer processors generate one or more balanced training sets based on one or more calculated training set statistics. The one or more computer processors train one or more cognitive models utilizing one or more unbalanced training sets and one or more generated balanced training sets. The one or more computer processors determine a fitness of the one or more trained cognitive models. The one or more computer processors adjust one or more training sets based on the determined fitness of the one or more cognitive models.