Data-Driven Model Configuration via Spectral Binning Preprocessing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data-driven models often fail to achieve sufficient prediction accuracy due to limited training data volumes, as they rely on prior knowledge of spectral properties for classification, which is not available for unknown classes.
Innovation Solution
A method involving preprocessing of measurement series through binning and additional steps like rescaling, normalization, and derivative calculation to enhance data quality and information content, allowing for effective learning even with limited data, and using neural networks or other models like support vector machines for classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data-driven models are trained using conventional methods with limited training data, then training time and computational resources are reduced, but prediction accuracy deteriorates
Solution Approach 1:
The patent applies preliminary action by performing preprocessing operations on training data before model training. This includes binning spectral data into bins with representative values, rescaling to normalize ranges, and handling missing values. These preprocessing steps are executed beforehand to enhance data quality and structure, enabling the model to achieve high prediction accuracy even with limited training samples by optimizing the input data characteristics prior to the learning process
Solution Approach 2:
The patent employs parameter changes by transforming the original spectral data through multiple preprocessing operations that modify data parameters. Binning converts continuous spectral values into discrete bin representations, rescaling adjusts value ranges to standardized intervals, and imputation replaces missing values with estimated parameters. These parameter transformations enhance the information content and quality of limited training data, allowing the model to learn more effectively from smaller datasets
2Reliability
If prior knowledge of spectral properties is used for classification, then classification accuracy is improved, but applicability to unknown classes deteriorates
Solution Approach 1:
The patent applies self-service by enabling the data-driven model to automatically learn classification patterns from preprocessed training data without requiring pre-defined spectral knowledge or manual feature engineering. The model autonomously identifies relationships between spectral characteristics and class labels during training, then applies this learned knowledge to classify both known and unknown samples. This self-learning capability allows the system to achieve accurate classification while maintaining adaptability to new, previously unknown classes
3Measurement precision
If measurement series are preprocessed through binning and additional steps, then data quality and information content are improved, but processing complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the preprocessing workflow into distinct, modular steps: binning spectral data into discrete intervals, rescaling to standardized ranges, handling missing values through imputation, and calculating derivatives. Each step processes the data independently and can be implemented as a separate computational module. This segmentation of the preprocessing complexity into manageable, sequential operations makes the enhanced data quality process systematic and implementable despite the multiple transformation steps involved
Data Source
AI summary
A method for the computer-assisted configuration of a data-driven model on the basis of training data is provided. The method is characterised in that the series of measurements are subjected to a suitable preprocessing process comprising a binning step, wherein measurement characteristics which existed during the measurement of the measurement values in question are taken into consideration. A suitable data-driven model such as a neural network is then learned on the basis of the pre-processed series of measurements. This learned data-driven model makes it possible to accurately forecast target vectors in accordance with associated series of measurements. The method can, for example, be used to analyse optical spectra. More particularly, it is possible to predict using the learned model whether the tissue sample for which an optical spectrum was detected represents diseased tissue.