Machine learning algorithms in microbial classification: a comparative analysis.

Area of Science:

• Biotechnology and Biomedical Engineering
• Infectious Disease Research
• Computational Biology and Bioinformatics

Background:

• Machine learning (ML) and deep learning (DL) are increasingly vital in healthcare, particularly for infectious disease prevention.
• Convolutional Neural Networks (CNNs) dominate image classification due to automated feature extraction.
• Transfer learning with pre-trained models addresses DL's data demands for microbial identification.

Purpose of the Study:

• To comparatively assess popular pre-trained CNN architectures for bacterial species classification.
• To evaluate the efficacy of transfer learning on a modest dataset for microbial identification.
• To identify the optimal CNN model for enhancing healthcare diagnostics and infectious disease prevention.

Main Methods:

• A comprehensive literature review of ML/DL in microbial diagnosis.
• Application of data augmentation to a dataset of ~660 bacterial images (33 species).
• Comparative evaluation of AlexNet, VGGNet, Inception, ResNet, and DenseNet-121 models.

Main Results:

• DenseNet-121 demonstrated superior performance in bacterial classification.
• Achieved peak accuracy of 99.08%, precision of 99.06%, recall of 99.00%, and F1-score of 98.99%.
• Transfer learning proved effective, mitigating the need for extensive training data.

Conclusions:

• DenseNet-121 is highly proficient for precise and efficient microbial identification using transfer learning.
• The findings support the integration of ML/DL in healthcare for improved diagnostics.
• This research contributes to advancing infectious disease prevention strategies through advanced computational methods.