Neural Network Feature Vector Generation for Prediction Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating prediction results from time series data fail to adequately reflect the characteristics of static data, leading to inaccurate predictions when combined with dynamic data in artificial neural network models.
Innovation Solution
A method that generates integrated and dynamic feature vectors from static and dynamic data using an artificial neural network model, where static data is categorized, grouped, and combined with dynamic data to create an adjacent matrix, enhancing the reflection of static data characteristics in prediction results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If simple embedding method is used to convert static data and combine with dynamic data, then the process is simple and fast, but the characteristics of static data cannot be reflected well leading to inaccurate predictions
Solution Approach 1:
The patent segments the feature extraction process into two distinct pathways: one for static data that incorporates group information and adjacent matrix computation, and another for dynamic data. This segmentation allows each data type to be processed with methods tailored to its characteristics, improving overall prediction accuracy without excessively complicating the entire system.
Solution Approach 2:
The patent introduces an adjacent matrix as an intermediary structure that captures relationships between different static data groups. This adjacent matrix serves as a mediator that enriches the static data representation before it is combined with dynamic data, enabling better reflection of static data characteristics while maintaining a structured approach to data integration.
2Loss of information
If static data is categorized and grouped with adjacent matrix computation, then the characteristics of static data are reflected better, but the processing time and computational complexity increase
Solution Approach 1:
The patent performs preliminary categorization and grouping of static data into meaningful groups before the main prediction process. By pre-computing the adjacent matrix that represents relationships between these groups, the system prepares enriched static data representations in advance, reducing the computational burden during real-time prediction while preserving static data characteristics.
Data Source
AI summary
Disclosed is a method for generating a prediction result by using static data and dynamic data according to an exemplary embodiment of the present disclosure. Specifically, according to the present disclosure, a computing device generates an integrated feature vector from static data and dynamic data of input data by using an artificial neural network model. The computing device generates a dynamic feature vector from the dynamic data of the input data by using the artificial neural network model. The computing device generates a final prediction result of the artificial neural network model based on the integrated feature vector and the dynamic feature vector.