Stochastic Subsampling for Imbalanced Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional digital content interaction prediction techniques are inaccurate when dealing with imbalanced classes in training data, leading to inefficiencies in user interaction control and missed conversion opportunities due to the skewing of prediction results by the majority class.
Innovation Solution
The proposed solution involves iterative sampling and processing of training data to ensure balanced representation of both positive and negative classes within batches, using stochastic subsampling gradient descent to train a model that predicts user interaction with digital content, thereby reducing variance and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If conventional prediction techniques use all training data including imbalanced classes, then the model can be trained on maximum available data, but the prediction accuracy deteriorates due to skewing by the majority class
Solution Approach 1:
The patent changes the parameter of class distribution in training batches by applying stochastic subsampling to the majority class, transforming the imbalanced training data into balanced batches while maintaining the overall training dataset size. This allows the model to learn from all available data without being skewed by class imbalance.
2Productivity
If the training data includes imbalanced classes, then all available training examples are utilized, but the prediction results become skewed toward the majority class
Solution Approach 1:
The patent introduces dynamic batch composition during training, where the majority class samples are stochastically subsampled in each training iteration to create dynamically balanced batches. This dynamic approach maintains training efficiency while improving prediction reliability by preventing majority class dominance.
3Measurement precision
If pre-subsampling is used to balance classes before training, then the training batches become balanced, but the process becomes more complex and time-consuming
Solution Approach 1:
The patent enables the training process to self-balance the classes through stochastic subsampling of the majority class during each training iteration. The system automatically adjusts batch composition without requiring external pre-processing intervention, simplifying the overall workflow while maintaining prediction accuracy.
Data Source
AI summary
Digital content interaction prediction and training techniques that address imbalanced classes are described. In one or more implementations, a digital medium environment is described to predict user interaction with digital content that addresses an imbalance of numbers included in first and second classes in training data used to train a model using machine learning. The training data is received that describes the first class and the second class. A model is trained using machine learning. The training includes sampling the training data to include at least one subset of the training data from the first class and at least one subset of the training data from the second class. Iterative selections are made of a batch from the sampled training data. The iteratively selected batches are iteratively processed by a classifier implemented using machine learning to train the model.