Clustered Flight Data Modeling for Higher Prediction Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning models for estimating or predicting target data from non-target data in aircraft flight data often result in insufficient estimation or prediction accuracy.
Innovation Solution
An information processing method that divides observation data into training and test sets, selects non-target data, classifies it into clusters based on density thresholds, generates models using non-linear regression, and iteratively refines parameter combinations and clustering settings to improve prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional machine learning models are used to estimate or predict target data from non-target data in aircraft flight data, then the model can be trained and deployed, but the estimation or prediction accuracy is insufficient
Solution Approach 1:
The patent divides the training data set into multiple clusters based on data density, creating separate models for different data regions. This segmentation allows each model to specialize in predicting target data for specific non-target data patterns, thereby improving overall estimation accuracy compared to a single conventional model
Solution Approach 2:
The patent generates different models for different clusters of non-target data, where each model is optimized for its specific cluster's characteristics. This local optimization ensures that predictions are made with higher accuracy for each specific data pattern rather than using a generic model for all data
2Measurement precision
If data is classified into clusters with density thresholds, then models can be generated for each cluster to improve accuracy, but the computational complexity and processing time increase
Solution Approach 1:
The patent changes the parameter of data density threshold to automatically classify non-target data into different clusters. By adjusting this parameter, the system can control the number and characteristics of clusters, balancing between model accuracy and computational complexity in generating multiple specialized models
Data Source
AI summary
In an information processing method of an embodiment, a computer is configured to divide observation data into training data and test data, select non-target data from the training data, select a parameter combination from a parameter set corresponding to the non-target data, classify the non-target data as clusters in a feature space of the parameter combination under clustering setting conditions, generate, for each of the clusters, a model trained to output target data included in the training data when the non-target data of the clusters is input, and reselect the parameter combination or change the clustering setting conditions such that a difference between the target data output from the model and the target data of the test data decreases.