Semi-Supervised Recurring Event Prediction via Fourier Transform

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional statistical techniques for predicting recurring events, such as appointments and transactions, are limited in accuracy and convert the problem into an unsupervised classification issue, where truth labels are unknown, making it difficult to effectively train machine learning models for predicting recurring events.

Innovation Solution

The method involves partitioning historical event timing data into two periods, predicting events in the second period using Fourier transforms, determining accurate predictions, and generating training data with heuristic filters to convert the problem into a semi-supervised classification, improving prediction accuracy by using machine learning models trained on this data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional statistical techniques are used to predict recurring events, then the problem can be solved without machine learning models, but the accuracy of prediction is limited and the problem becomes an unsupervised classification issue where truth labels are unknown

Engineering Contradiction:
Improveprediction accuracyVSAvoidclassification problem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces an intermediary process that uses Fourier transforms to generate predicted event labels from historical timing data. These predicted labels serve as a mediator between the raw unsupervised data and the machine learning model, converting the unsupervised classification problem into a semi-supervised one where truth labels are now available for training

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent changes the parameter state of the classification problem by transforming the unknown truth labels into known predicted labels through Fourier analysis. This parameter change enables the use of supervised learning techniques while maintaining the underlying statistical patterns in the event timing data

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If machine learning models are trained to identify recurring events, then prediction accuracy can be improved, but the problem requires converting unsupervised classification into semi-supervised classification which increases system complexity

Engineering Contradiction:
Improveprediction accuracyVSAvoidclassification system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent performs preliminary action by generating predicted event labels using Fourier transforms before training the machine learning model. This preliminary labeling process prepares the data in advance, converting the unsupervised problem into a semi-supervised one that can be effectively solved by machine learning algorithms

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent substitutes the mechanical process of manual label creation with an automated Fourier transform-based prediction system. This substitution efficiently generates large amounts of training data with predicted labels, enabling machine learning model training without manual intervention while improving prediction accuracy

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Quantity of substance

If Fourier transforms are used to generate predicted events for training data, then training data can be generated with predicted labels, but the process requires partitioning historical data into time periods and comparing predictions to actual events

Engineering Contradiction:
Improvetraining data quantityVSAvoiddata processing complexity
Core Design Contradiction:
Quantity of substanceVSDevice complexity

Solution Approach 1:

The patent segments historical event timing data into distinct time periods (first time period for training, second time period for validation). This segmentation enables the generation of training data by predicting events in one period and validating against actual events in another period, creating structured training datasets

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent employs periodic action by using Fourier transforms to identify recurring patterns at different frequencies in the event timing data. This periodic analysis enables the generation of predicted labels that capture the cyclic nature of recurring events, providing rich training signals for the machine learning model

Inventive Principle:
Principle #19Periodic action

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

This approach enhances the accuracy of predicting recurring events by converting the unsupervised classification problem into a semi-supervised one, allowing machine learning models to better identify recurring series of events, improving forecasting beyond conventional statistical techniques.

Implementation Method 1

predicting, based on the first portion of the historical event timing data, a first plurality of predicted events corresponding to the second time period

Methodology Applied
Scientific EffectFourier transform:

Data Source

PatentEP4428770A1Identifying recurring events using automated semi-supervised classifiers
Publication Date: 2024.09.11 INTUIT INC
  • EP4428770A1 patent drawingFigure 1
  • EP4428770A1 patent drawingFigure 2
  • EP4428770A1 patent drawingFigure 3

AI summary

Systems and methods for training machine learning models are disclosed. An example method includes receiving historical event timing data including event data for a first portion including events from a first time period, and a second portion comprising events from a second time period not including the first time period, predicting, based on the first portion of the historical event timing data, a first plurality of predicted events, the first plurality of predicted events corresponding to the second time period, determining a first subset of predicted events to be accurate predictions based at least in part on comparing the first plurality of predicted events to the historical events occurring within the second time period, generating training data based at least in part on the first subset of the first plurality of predicted events, and training the machine learning model based at least in part on the training data.