Sensitivity and generalized analytical sensitivity expressions for quantitative analysis using convolutional neural...

Area of Science:

• Analytical Chemistry
• Chemometrics
• Machine Learning

Background:

• Deep neural networks (DNNs) and convolutional neural networks (CNNs) are increasingly used in analytical chemistry for calibration.
• Existing performance measures for these networks primarily focus on prediction error, limiting objective model comparison.
• There is a need for comprehensive figures of merit to evaluate the analytical properties of neural networks.

Purpose of the Study:

• To propose and validate a novel method for calculating generalized analytical sensitivity for neural networks.
• To define and compute this sensitivity as an additional figure of merit for calibration models.
• To investigate the impact of regularization dataset size on CNNs and compare them with conventional methods.

Main Methods:

• Development of a generalized analytical sensitivity calculation for neural networks.
• Application of the method to both simulated and real-world analytical datasets.
• Comparative analysis of CNNs, considering regularization dataset size, against traditional calibration techniques.

Main Results:

• The proposed method provides a new, objective figure of merit for neural network calibration models.
• Generalized analytical sensitivity can be effectively calculated for various neural network architectures.
• The study elucidates the relationship between CNN performance, regularization, and dataset size.

Conclusions:

• The generalized analytical sensitivity offers a more complete assessment of neural network performance in analytical chemistry.
• This metric facilitates more robust and objective comparisons between different deep learning models.
• Understanding the influence of regularization and dataset size is crucial for optimizing CNNs in analytical applications.