Time-Series Imputation via Joint Left-Right Value Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current time-series prediction techniques using Recurrent Neural Networks (RNNs) for missing data scenarios primarily consider the last observed value from the left, ignoring the immediate observed value from the right, leading to errors in imputation and forecasting.
Innovation Solution
A processor-implemented method and system that uses a joint impute and learn technique, selecting both the closest left and right data values, along with a mean value, to calculate a new data value for missing entries in time-series data, and passes these to a SoftMax layer to determine a probability distribution for accurate imputation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If only the last observed value from the left is used for imputation, then the imputation process is simple, but the prediction accuracy deteriorates due to ignoring useful right-side information
Solution Approach 1:
The patent merges multiple imputation approaches (left-side last observed value, right-side immediate observed value, and mean-based imputation) into a unified framework. The system combines these different sources of information to fill missing values, thereby improving prediction accuracy while maintaining reasonable computational efficiency.
Solution Approach 2:
The patent dynamically selects the most appropriate imputation strategy based on the specific context of each missing value. Instead of using a static single-method approach, the system adapts by evaluating multiple options (left, right, mean) and selecting or combining them based on their suitability for the particular data pattern and missingness scenario.
2Device complexity
If traditional RNN-based prediction is used with missing data, then the model structure remains simple, but forecasting reliability deteriorates due to improper handling of missing entries
Solution Approach 1:
The patent applies preliminary action by performing data imputation before the RNN-based prediction process. Missing values are filled using the joint imputation technique (combining left, right, and mean approaches) prior to feeding data into the RNN model, ensuring that the model receives complete and reliable input sequences, thereby improving forecasting reliability without complicating the RNN structure itself.
Solution Approach 2:
The patent introduces an intermediary imputation layer between the raw data and the RNN model. This intermediary component processes missing values using multiple strategies (left-side observation, right-side observation, mean calculation) and outputs completed data sequences to the RNN, thereby protecting the model from the adverse effects of missing data while maintaining the original model architecture.
3Productivity
If immediate observed value from the right is ignored, then the data processing is faster, but imputation accuracy deteriorates due to loss of useful information
Solution Approach 1:
The patent merges multiple imputation sources including the previously ignored right-side immediate observed value. By combining left-side last observed value, right-side immediate observed value, and mean-based imputation, the system recovers useful information that would otherwise be lost, improving imputation accuracy while maintaining efficient processing through parallel computation of these multiple sources.
Data Source
AI summary
Currently available time-series prediction techniques only factors last observed value from left of missing values and immediate observed value from right is mostly ignored while performing data imputation, thus causing errors in imputation and learning. Present application provides methods and systems for time-series prediction under missing data scenarios. The system first determines missing data values in time-series data. Thereafter, system identifies left data value, right data value, left gap length, right gap length and mean value for each missing data value. Further, system provides left gap length and right gap length identified for each missing data value to feed-forward neural network to obtain importance of left data value, right data value and mean value. The system then passes importance obtained for each missing data value to SoftMax layer to obtain probability distribution that is further utilized to calculate new data value corresponding to each missing data value in time-series data.