Intelligent Discrete Deep Learning Based Classification Methodology in Chemometrics.

Area of Science:

• Chemometrics
• Machine Learning
• Artificial Intelligence

Background:

• Deep learning models are popular in chemometrics for classification due to their accuracy and ability to model complex patterns.
• Current deep learning classifiers use continuous cost functions, which can limit performance on discrete classification tasks.

Purpose of the Study:

• To propose and implement a novel discrete learning-based classification approach for deep feed-forward neural networks.
• To address the conflict between continuous cost functions and the discrete nature of classification problems in chemometrics.

Main Methods:

• Developed a new learning process based on maximizing a discrete matching function instead of minimizing a continuous distance-based cost function.
• Implemented the discrete learning approach on a deep feed-forward neural network.
• Evaluated the approach on five benchmark chemistry datasets.

Main Results:

• The proposed discrete deep learning approach demonstrated superior performance compared to its classic continuous counterpart.
• Empirical results confirmed the significant impact of discrete learning processes on deep learning classification model performance.
• The discrete approach showed improved accuracy in analyzing chemical data.

Conclusions:

• Discrete learning processes are crucial for enhancing the performance of deep learning classification models in chemometrics.
• The proposed discrete deep learning methodology offers a powerful alternative for chemical data analysis.
• This research highlights the importance of aligning learning algorithms with the inherent data properties for better classification outcomes.