Time Series Prediction Model for Payment Interaction Metrics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems face challenges in estimating interaction metrics in payment processing networks due to concept drift, multi-modality, and large interaction datasets, which affect real-time monitoring and prediction accuracy.
Innovation Solution
A method using an interaction encoder and temporal encoder to generate hidden representations from interaction data, combined with a scale and shape decoder to predict multi-horizon interaction metrics, enabling efficient estimation and handling of concept drift and multi-modality through online learning and data sampling techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional time series prediction methods are used, then the system can handle basic prediction tasks, but the accuracy deteriorates due to concept drift and multi-modality in payment data
Solution Approach 1:
The patent applies dynamics by making the prediction model adaptive through online learning. The system continuously updates its parameters based on incoming payment data, allowing it to adapt to concept drift and multi-modality. The model transitions from a static structure to a dynamic one that evolves with the data distribution changes in the payment network.
Solution Approach 2:
The patent changes parameters by transforming the prediction approach from direct value prediction to predicting scale and shape parameters of a distribution. This parameter transformation allows the model to capture multi-modality better, as different modes can be represented as variations in scale and shape parameters rather than requiring the model to predict each mode separately.
2Productivity
If the system processes large interaction datasets in real-time, then monitoring capability is improved, but computational resources and processing time are overwhelmed
Solution Approach 1:
The patent extracts only the essential features from the large interaction datasets by using embedding layers to represent entities and their relationships in a compressed form. Instead of processing raw transaction data directly, the system extracts key interaction patterns and represents them in a lower-dimensional space, significantly reducing computational requirements while maintaining predictive accuracy.
Solution Approach 2:
The patent performs preliminary action by pre-computing entity embeddings and interaction representations before the actual prediction task. The system prepares the data in advance by encoding entity relationships and temporal patterns, so that during real-time operation, only the final prediction step is needed, dramatically reducing online processing time.
3Measurement precision
If the model captures detailed time-varying patterns, then prediction accuracy is improved, but model complexity and computational burden increase
Solution Approach 1:
The patent segments the prediction task into two independent components: scale prediction and shape prediction. By dividing the complex prediction problem into separate sub-tasks, the model can capture detailed time-varying patterns for each aspect without requiring a single overly complex model. This segmentation reduces overall model complexity while maintaining the ability to capture intricate temporal patterns.
Data Source
AI summary
A method performed by a computer is disclosed. The method comprises receiving interaction data between electronic devices of a plurality of entities. The interaction data is used to form an entity interaction vector containing a number of interactions between the electronic devices of a chosen entity and an entity time series containing a plurality of metrics per unit time of the interactions. An interaction encoder of the computer can generate an interaction hidden representation of the entity interaction vector using embeddings of the plurality of entities. A temporal encoder of the computer can generate a temporal hidden representation of the entity time series. The interaction hidden representation and the temporal hidden representation can be used to generate a predicted scale and a shape estimation of a target interaction metric. The computer can then generate an estimated interaction metric of a time period using the predicted scale and the shape estimation.