Successive Feature Imputation Using Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional methods for filling missing feature values in datasets often overlook valuable information from other features, leading to biases and analysis errors in machine learning models, as they either eliminate incomplete records or use simplistic imputation strategies.
Innovation Solution
A method using machine learning to recursively impute missing feature values by sorting features by their missing value count, determining the data type, selecting appropriate models, training them using populated indices, and predicting missing values, thereby leveraging information across features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional imputation methods eliminate incomplete rows or use simple mean/median replacement, then the imputation process is simple and fast, but the accuracy of imputation decreases and information from other features is lost
Solution Approach 1:
The patent segments the imputation process into multiple iterations, where in each iteration a machine learning model is trained on features with fewer missing values and used to impute features with more missing values. This sequential segmentation allows the system to leverage information from partially complete records while maintaining manageable computational complexity at each step.
Solution Approach 2:
The patent performs preliminary actions by first identifying and imputing features with the lowest missing value counts before proceeding to features with higher missing value counts. This preliminary imputation of easier-to-fill features creates a foundation of more complete data that can be used to inform subsequent imputation of harder-to-fill features, improving overall accuracy.
2Reliability
If traditional methods use mean or median of populated values, then the imputation process is simple, but biases and analysis errors increase in machine learning models
Solution Approach 1:
The patent introduces machine learning models as intermediaries between the raw data with missing values and the final imputed dataset. These models act as mediators that can capture complex relationships and dependencies between features, producing more reliable imputations than simple statistical measures while maintaining a structured processing pipeline.
Solution Approach 2:
The patent replaces the mechanical, deterministic approach of mean/median calculation with a learning-based system that can adapt to the specific patterns in the data. This substitution allows the system to learn complex relationships between features and produce more reliable imputations that account for contextual dependencies.
3Loss of information
If machine learning models are trained on partially filled datasets, then useful information from other features can be utilized, but the complexity of the imputation process increases
Solution Approach 1:
The patent employs periodic action by iteratively training machine learning models and performing imputation in cycles. Each iteration processes a subset of features and updates the dataset, allowing information to be progressively incorporated. This periodic approach balances information retention with manageable processing time by breaking down the complex task into repeated, smaller steps.
Data Source
AI summary
Systems and methods for successively imputing missing feature values using machine learning to sequentially fill in missing feature values in partially-filled datasets, and by using the information in populated records of the dataset. The systems and methods disclosed herein may be useful in many machine learning contexts and application where datasets are missing values.