Time Series Prediction via Classifier-Regressor Scaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting characteristics of specific data entries within large datasets are computationally slow and prone to inaccurate predictions, especially when the specific data entries are a small portion of the larger dataset.
Innovation Solution
A combined classifier-regressor prediction processor is used, which includes an XGBoost classifier and a regressor mechanism employing an attention mechanism and a Long Short-Term Memory (LSTM) neural network. The classifier is trained on the complete dataset, while the regressor is trained on a pruned dataset with zeroes removed, and the output of the regressor is scaled by the classifier output to adjust for overfitting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a conventional processor applies data filters based on trackable parameters to predict characteristics of specific data entries, then the prediction can be obtained, but the computation becomes slow and predictions become inaccurate
Solution Approach 1:
The patent segments the prediction task into two distinct stages: (1) a classification stage that identifies relevant data entries using a trained classifier, and (2) a regression stage that predicts characteristics using a trained regressor. This segmentation allows each stage to specialize in its function, improving both accuracy and efficiency compared to a single conventional filtering approach
Solution Approach 2:
The patent performs preliminary classification to identify and filter relevant data entries before performing the regression prediction. By pre-classifying which data entries require prediction, the system avoids unnecessary computation on irrelevant entries, thereby improving computational speed while maintaining prediction accuracy for the target subset
2Adaptability or versatility
If the specific set of data entries is a small portion of the large data set and is not easily identifiable, then prediction becomes more challenging, but conventional methods still struggle with both speed and accuracy
Solution Approach 1:
The system segments the data processing into classification (identifying the small subset) and regression (predicting characteristics) phases. The classifier efficiently identifies the small portion of relevant data entries, and only then does the regressor perform computation on this reduced subset, significantly reducing computational time while maintaining adaptability to various data scenarios
Solution Approach 2:
The classification model acts as an intermediary between the large dataset and the prediction process. It filters and selects the small relevant subset that needs prediction, serving as a mediator that reduces the problem size before the regressor processes the data, thereby reducing computational time while maintaining accuracy
Data Source
AI summary
Systems and methods for computing time series predictions are provided, including a two-stage classifier-and-regressor processor. The classifier is trained on the complete data set while the regressor is trained on a pruned dataset. The classifier includes an Extreme Gradient Boosting classifier. The regressor includes an attention mechanism and a Long Short-Term Memory (LSTM) neural network. For a series of successive time period computations, a current output of the LSTM neural network is recursively fed back as an input to the attention mechanism for a subsequent time period computation. The output of the regressor is scaled by the output of the classifier to adjust for overfitting caused by the pruned training dataset.