Hybrid Machine Learning Models for Payment Transaction Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Predicting future states based on time series data is challenging due to difficulties in selecting relevant features for machine learning models, leading to inefficiencies and inaccuracies, especially when dealing with large volumes of data from payment transactions.
Innovation Solution
A method using feature engineering and hybrid machine learning models, specifically involving a random forest model for feature ranking, followed by inputting these features into a denoising autoencoder, convolutional neural network, and long short-term memory model to predict net settlement positions, enhancing prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If too many features are used for machine learning models, then the model may capture more patterns, but computational resources are wasted especially when dealing with large volumes of payment transaction data
Solution Approach 1:
The patent extracts and selects only the most relevant features from a large set of potential features using feature selection techniques. This involves identifying and retaining key features that have the highest impact on prediction accuracy while discarding redundant or less important features, thereby reducing computational resource consumption while maintaining predictive performance.
Solution Approach 2:
The patent applies partial action by using a subset of features rather than all available features. Feature selection methods identify the optimal number and type of features needed for accurate predictions, avoiding the excessive use of computational resources that would result from processing all possible features.
2Productivity
If features that are poorly correlated with the target prediction are used, then more data can be processed, but prediction accuracy deteriorates
Solution Approach 1:
The patent changes the parameters of feature selection by evaluating features based on their correlation with target predictions. Feature importance metrics and selection algorithms identify which features have strong correlations with the prediction target, allowing the system to process only those features that contribute meaningfully to prediction accuracy.
3Device complexity
If a single type of machine learning model is selected, then the model is simpler to implement, but prediction accuracy suffers because the model may not learn certain types of patterns better than others
Solution Approach 1:
The patent merges multiple machine learning models into a hybrid ensemble system. Different models (e.g., tree-based models, neural networks, probabilistic models) are combined to leverage their respective strengths in learning different types of patterns. The ensemble approach integrates predictions from multiple models to achieve superior overall accuracy compared to any single model.
Solution Approach 2:
The patent creates a composite modeling approach by combining different types of machine learning models into a hybrid system. Each model type contributes unique capabilities for capturing different patterns in the data, similar to how composite materials combine different substances to achieve superior properties. This composite model structure enables the system to learn diverse patterns that would be difficult for any single model type to capture alone.
Data Source
AI summary
Provided is a method for predicting future states based on time series data using feature engineering and/or hybrid machine learning models. The method may include receiving payment transaction data associated with a plurality of payment transactions, the plurality of payment transactions including a first subset of payment transactions associated with a first entity; determining a plurality of features based on the payment transaction data associated with the plurality of payment transactions; inputting the plurality of features into at least one machine learning model to provide at least one prediction of a net settlement position of the first entity; and communicating the at least one prediction of the net settlement position to a first entity system associated with the first entity. A system and computer program product are also disclosed.