Imputing Machine Learning Training Data via Cluster Regression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning training data often contains missing values, which can decrease the effectiveness of training models when not handled properly, as removing these values can lead to the loss of useful data points and existing imputation methods may not provide relevant replacements.
Innovation Solution
A method that determines a correlation list of missing value predictors, generates a cluster model based on target and predictor values, and uses a linear regression model to iteratively impute missing values within clusters, ensuring that imputed values are relevant and accurate.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If missing values are removed from training data, then model training effectiveness is improved by eliminating incomplete data, but data loss occurs as useful data points are discarded
Solution Approach 1:
The patent extracts and identifies missing value predictors from the training data, separating the imputation task from the main training process. By extracting the imputation sub-task and handling it separately through correlation analysis and cluster modeling, the system preserves complete data rows while ensuring quality imputed values, thus avoiding data loss while maintaining training effectiveness
Solution Approach 2:
The patent changes the state of missing values by transforming them from gaps to imputed values through a multi-step process involving correlation calculation, cluster assignment, and regression-based imputation. This parameter transformation allows the system to retain all data points while converting incomplete information into useful training data, resolving the contradiction between data retention and training quality
2Quantity of substance
If existing imputation methods are used, then missing values are replaced to maintain data completeness, but imputed values may not be relevant leading to reduced model accuracy
Solution Approach 1:
The patent applies local quality by creating context-specific imputation for each missing value based on its local data environment. Through correlation analysis with other predictors and assignment to specific clusters, each imputed value is tailored to its local context rather than applying a global imputation rule. This ensures that imputed values are relevant to the specific row and cluster characteristics, maintaining both data completeness and imputation accuracy
Solution Approach 2:
The patent introduces cluster models as intermediary structures between the raw data and imputed values. These clusters act as mediators that capture local data patterns and characteristics, allowing the imputation process to leverage cluster-specific information. The intermediary cluster structure enables more accurate imputation by providing contextual information that bridges the gap between incomplete data and meaningful replacements
3Measurement precision
If cluster model with linear regression is used for imputation, then imputation accuracy is improved by using context-specific models, but computational complexity increases
Solution Approach 1:
The patent segments the training data into multiple clusters based on predictor correlations and data characteristics. By dividing the data into smaller, more homogeneous groups, the system can apply linear regression models to each cluster separately, improving imputation accuracy through context-specific modeling. The segmentation approach manages computational complexity by working with smaller subsets of data rather than the entire dataset at once
Data Source
AI summary
Embodiments are disclosed for a method. The method includes determining a correlation list of missing value predictors. The method also includes generating a cluster model having multiple clusters. The cluster model is based on a target value and predictor values. The method further includes determining an imputed value for a missing value of a row of original training data based on a linear regression model for multiple non-missing value predictor values for the clusters.