Multi-Stream RNN for Event Prediction Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for multivariate event prediction, such as fraud detection and automatic network communication, face challenges in accurately modeling and predicting event behavior over future time periods, leading to inefficiencies and resource wastage due to the need for broader predictions.
Innovation Solution
A computer-implemented method and system using multi-stream recurrent neural networks (RNNs) that aggregate event data at a fine granularity, employing stacked and shrinking RNN models to generate feature vectors and predict future events, thereby improving prediction accuracy and resource allocation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If broader predictions are made to cover more event scenarios, then prediction coverage is improved, but system efficiency and resource allocation deteriorate
Solution Approach 1:
The patent applies local quality by segmenting the prediction task into multiple specialized RNN models, each dedicated to predicting specific event types or patterns. Instead of using a single broad prediction system that attempts to handle all scenarios, the invention creates focused models that specialize in specific temporal patterns (e.g., hourly, daily, weekly cycles), thereby improving both prediction coverage for specific scenarios and system efficiency through targeted processing.
Solution Approach 2:
The patent segments the prediction system into multiple independent RNN models, each responsible for specific event prediction tasks. This segmentation allows the system to handle diverse prediction scenarios through separate specialized models rather than a monolithic broad prediction system, improving both adaptability to different event types and overall system efficiency through parallel processing.
2Adaptability or versatility
If historical event data is used to predict future patterns, then prediction capability is improved, but measurement precision and accuracy deteriorate
Solution Approach 1:
The patent applies parameter changes by transforming historical event data into feature vectors with specific temporal aggregations (hourly, daily, weekly summaries). The system adjusts the granularity and transformation parameters of the data to optimize prediction accuracy for different time horizons and event types, thereby improving measurement precision while maintaining broad prediction capability.
Solution Approach 2:
The patent segments the prediction capability into multiple specialized RNN models, each trained on and optimized for specific temporal patterns and event characteristics. This segmentation allows each model to achieve high precision for its specialized task while the collective system maintains broad prediction capability across different scenarios.
3Measurement precision
If multi-stream RNN models are used to aggregate event data at fine granularity, then prediction accuracy is improved, but device complexity increases
Solution Approach 1:
The patent segments the complex prediction task into multiple simpler specialized RNN models, each handling specific temporal patterns. This segmentation reduces the complexity of individual models compared to a single monolithic model that would need to handle all scenarios, while the collection of specialized models achieves high overall accuracy through their individual strengths.
Solution Approach 2:
The patent creates a universal prediction framework where multiple RNN models can be applied to different event types and time horizons through a common architecture and data processing pipeline. This multi-functionality allows the system to achieve high accuracy across diverse scenarios while managing complexity through code reusability and standardized processing steps.
Data Source
AI summary
Described are a system, method, and computer program product for multivariate event prediction using multi-stream recurrent neural networks. The method includes receiving event data from a sample time period and generating feature vectors for each subperiod of each day. The method also includes providing the feature vectors as inputs to a set of first recurrent neural network (RNN) models and generating first outputs for each RNN node. The method further includes merging the first outputs for each same subperiod to form aggregated time-series layers. The method further includes providing the aggregated time-series layers as an input to a second RNN model and generating final outputs for each RNN node of the second RNN model.