Inspection Value Prediction Model With Interpretable Trend Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing systems for predicting future inspection values suffer from decreased interpretability due to varying tendencies in short-term and long-term predictions, making it difficult to understand the behavior of the learning model.

Innovation Solution

A learning model generation apparatus that performs machine learning using inspection values and function parameters to generate a model where the inspection value is the explanatory variable and the parameter is the objective variable, allowing for the prediction of future inspection values with improved interpretability.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Duration of action of moving object

If multi-stage prediction is performed (predicting 1 year, then 1-2 years, then 2-3 years sequentially), then future inspection values can be predicted beyond one year, but the tendency of change varies in each prediction stage reducing interpretability

Engineering Contradiction:
Improveprediction time horizonVSAvoidinterpretability
Core Design Contradiction:
Duration of action of moving objectVSLoss of information

Solution Approach 1:

The patent divides the prediction process into two distinct components: a tendency determination unit that identifies the overall direction (increasing, decreasing, or fluctuating) of inspection value changes, and a value calculation unit that computes specific predicted values. This segmentation allows the system to maintain consistent interpretability across different prediction time horizons while still providing long-term forecasts.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces a tendency determination unit as an intermediary component that bridges the gap between raw prediction data and interpretable results. This unit analyzes the overall trend of inspection value changes and provides a clear directional indicator, making the prediction results interpretable regardless of the prediction time horizon.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Loss of time

If the learning model predicts inspection values for multiple time periods, then long-term health trends can be identified, but the varying tendencies in different prediction stages make model behavior difficult to interpret

Engineering Contradiction:
Improveprediction time spanVSAvoidmodel interpretability
Core Design Contradiction:
Loss of timeVSEase of operation

Solution Approach 1:

The patent separates the prediction functionality into distinct modules: a tendency determination unit that handles interpretability by identifying overall trends, and a value calculation unit that handles specific predicted values. This segmentation allows the model to provide long-term predictions while maintaining clear interpretability through the separate tendency analysis.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Instead of directly outputting prediction values and hoping they are interpretable, the patent inverts the approach by first determining the tendency (interpretability) and then calculating the specific values. This inversion ensures that interpretability is established before numerical predictions are made, making model behavior easier to understand.

Inventive Principle:
Principle #13The other way round (Inversion)

Data Source

PatentEP4693331A1Learning model generation device, inspection value prediction device, learning model generation method, inspection value prediction method, and computer-readable recording medium
Publication Date: 2026.02.11 NEC SOLUTION INNOVATORS LTD
  • EP4693331A1 patent drawingFigure 1
  • EP4693331A1 patent drawingFigure 2
  • EP4693331A1 patent drawingFigure 3

AI summary

This learning model generation device 10 is equipped with a learning model generation unit 11 which, when a function expressing a change in an inspection value obtained by inspecting a person is set, generates a learning model in which the inspection value is the explanatory variable and the parameter is the objective variable, by performing machine learning using inspection values of sample people and parameters of the function for the sample people as training data.