Incremental Time Window Procedure for Training Sample Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional methods for training artificial intelligence models often introduce bias, leading to inaccurate predictions, particularly when analyzing recurring data patterns in transaction data, as they fail to detect irregular or complex patterns such as ceased or paused transactions.
Innovation Solution
An incremental time window method is employed to generate labels for training machine learning models, which involves grouping data points into analysis and holdout portions based on recurrence periods, allowing for the identification of input features and labels that help the models detect irregular patterns and avoid biases.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional methods are used to train AI models, then the training process is simple, but the model accuracy deteriorates due to bias and inability to detect irregular patterns
Solution Approach 1:
The training dataset is divided into multiple time windows, where each time window contains a subset of training samples. This segmentation allows the model to learn from different temporal segments independently, reducing bias from any single period while capturing irregular patterns like ceased or paused transactions that traditional single-batch training would miss.
Solution Approach 2:
The patent employs dynamic sampling strategies where the selection and weighting of training samples adapt based on temporal characteristics and pattern recognition. The training process dynamically adjusts which time windows and samples are emphasized, allowing the model to focus on irregular patterns while maintaining overall training efficiency.
2Difficulty of detecting and measuring
If traditional training methods are used, then training speed is fast, but the ability to detect irregular patterns deteriorates
Solution Approach 1:
The patent performs preliminary analysis to identify irregular patterns such as ceased, paused, or modified recurring transactions before model training. By pre-processing the data to flag these anomalies and incorporating them as specific training features, the model can detect irregular patterns more effectively without requiring exhaustive training on all possible pattern variations.
Solution Approach 2:
The training process utilizes multiple parameters including time window sizes, sampling rates, and pattern threshold values. By adjusting these parameters to optimize for irregular pattern detection, the system achieves better pattern recognition capability while managing training computational requirements through parameter optimization rather than brute-force approaches.
Data Source
AI summary
Disclosed herein are system, method, and computer program product embodiments for generating labels for training a machine learning mode using an incremental time window process. The described process may be used in a recurrence detection system. A dataset may be analyzed using incremental split dates to divide the dataset into an analysis portion and a holdout portion. The analysis portion may be analyzed to determine input features related to a predicted recurrence in the dataset. The holdout portion may be tested against the analysis portion and the input features to generate a label. The label may indicate whether or not the holdout portion confirms the prediction. The testing of the holdout portion against the analysis portion may be repeated by incrementally using different split dates and multiple separate analysis portions and holdout portions to generate multiple labels and corresponding input features.