Scholarlink

This paper evaluates the performance and relevance of classical machine learning algorithm—Multinomial Naive Bayes (MNB) classifier with Term Frequency Inverse Document Frequency (TF-IDF) vectorizer, in diagnosing and classifying diseases based on natural language symptoms narratives of the patients. In this study, we highlighted the advantages of MNB—such as computational efficiency, interpretability, and scalability—by comparing its performance with existing deep-learning approaches including Medical Concept Normalization- Bidirectional Encoder Representations from Transformers (MCN-BERT) models optimized with AdamP and AdamW optimizers, and a Bidirectional Long Short-Term Memory (BiLSTM) model optimized with Hyperopt. The proposed model achieved an accuracy of 98.75%, an F1-Score of 98.74%, a recall of 98.75%, and a precision of 98.79%. The resulting training time for the proposed model is at least 50 times faster than the deep-learning approaches in the comparative study. This performance and significant reduction in computational demand and time complexity demonstrates the suitability of Bayesian classifiers, specifically MNB+TF-IDF, in medical text classification tasks as presented in this study.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Disease Diagnosis Through Patient-Reported Symptoms Narratives Using Multinomial Naive Bayes

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Edgardo R. Callagon,
  • Prospero C. Naval

摘要

This paper evaluates the performance and relevance of classical machine learning algorithm—Multinomial Naive Bayes (MNB) classifier with Term Frequency Inverse Document Frequency (TF-IDF) vectorizer, in diagnosing and classifying diseases based on natural language symptoms narratives of the patients. In this study, we highlighted the advantages of MNB—such as computational efficiency, interpretability, and scalability—by comparing its performance with existing deep-learning approaches including Medical Concept Normalization- Bidirectional Encoder Representations from Transformers (MCN-BERT) models optimized with AdamP and AdamW optimizers, and a Bidirectional Long Short-Term Memory (BiLSTM) model optimized with Hyperopt. The proposed model achieved an accuracy of 98.75%, an F1-Score of 98.74%, a recall of 98.75%, and a precision of 98.79%. The resulting training time for the proposed model is at least 50 times faster than the deep-learning approaches in the comparative study. This performance and significant reduction in computational demand and time complexity demonstrates the suitability of Bayesian classifiers, specifically MNB+TF-IDF, in medical text classification tasks as presented in this study.