USING MACHINE LEARNING ALGORITHMS TO PREDICT TRANSACTIONS THAT MATCH EACH OTHER USING PATTERNS FROM MATCHING FEEDBACK

Smart overview of the Invention

The patent application describes a system leveraging machine learning algorithms to match transactions across different systems by analyzing aggregate record data. This approach involves using preliminary matching tools to define rules for reconciling most transactions, while remaining transactions undergo an interactive matching process. The system can match transactions in various configurations, such as many-to-many or one-to-many, and presents potential matches with confidence levels and insights through a user interface.

Machine Learning Model

A hybrid machine learning model underpins the system, utilizing algorithms like random forests, decision trees, neural networks, and others. This model is trained on previous manual matches and continuously evolves by incorporating user feedback on suggested matches. The model's ability to adapt ensures that it remains effective in identifying accurate matches over time, enhancing its predictive capabilities.

Matching Process

The matching process involves displaying data from two tables, such as sales transactions and bank deposits, and generating vector embeddings for these data sets. By combining values from records and evaluating conditions, the system creates vector embeddings that represent the data. The model then calculates distances between these embeddings to determine the closeness of potential matches, applying different weights to various component-level differences.

Graphical User Interface

The system's graphical user interface visually represents confidence levels in potential matches through color-coded displays. This allows users to quickly assess which matches have higher confidence levels and make informed decisions. Users can choose between different sets of matched records, with their selections stored for easy retrieval and comparison, facilitating a streamlined matching process.

Practical Application

This technology addresses the challenge of transaction misalignment in organizations, which can lead to financial losses. By automating the matching process and providing insightful feedback, the system reduces the effort required to verify transaction alignment across different systems. Consequently, it helps organizations minimize financial discrepancies and improve the accuracy of their transaction tracking and reconciliation efforts.