FREGEX: A Feature Extraction Method for Biomedical Text Classification using Regular Expressions.

Area of Science:

• Computational Biology and Bioinformatics
• Natural Language Processing in Biomedicine
• Machine Learning for Health Informatics

Background:

• Biomedical text mining requires effective feature extraction methods.
• Traditional methods like n-grams can be computationally intensive and less precise.
• Automated feature engineering is crucial for scalable analysis of large biomedical corpora.

Purpose of the Study:

• To introduce FREGEX, a new method for automatic feature extraction from biomedical texts.
• To evaluate FREGEX's effectiveness compared to n-gram features using sequence alignment algorithms.
• To assess the impact of FREGEX on machine learning classifier performance for health-related datasets.

Main Methods:

• Developed FREGEX utilizing regular expressions for token pattern identification.
• Employed Smith-Waterman and Needleman-Wunsch algorithms for sequence alignment and feature extraction.
• Compared FREGEX-derived features against n-grams, both represented using TF-IDF vectors, and trained Support Vector Machine and Naïve Bayes classifiers.

Main Results:

• FREGEX significantly improved the performance of both Support Vector Machine and Naïve Bayes classifiers across all evaluated datasets.
• FREGEX utilized fewer features compared to n-grams, particularly for datasets involving anthropometric measures (obesity and obesity types).
• The method demonstrated superior feature extraction capabilities for biomedical text analysis.

Conclusions:

• FREGEX offers a more efficient and effective approach to feature extraction in biomedical natural language processing.
• The regular expression-based method enhances machine learning model accuracy and reduces feature dimensionality.
• FREGEX shows particular promise for analyzing datasets related to anthropometric and behavioral health information.