Scholarlink

<p>In natural language processing, text classification is a significant challenge, requiring effective word embedding techniques and machine learning models to achieve optimal results. In this paper, we propose a method to enhance machine learning models through the optimal selection of word embedding techniques and model hyperparameters using genetic algorithms. We explore three popular word embedding techniques, namely Word2Vec, GloVe, and BERT, and integrate them with four different machine learning algorithms for text classification. The genetic algorithm is used to simultaneously optimize word embedding techniques and machine learning hyperparameters, identifying the best combination for each classifier to achieve optimal accuracy. The proposed WET-ML-GA approach was tested on two medical datasets, Symptom2Disease and Mental Health. Experimental results demonstrate that genetic algorithms effectively select the optimal configurations, resulting in enhanced performance. In particular, for each dataset, our approach selects the best word embedding technique and the best hyperparameters for each classifier, based on accuracy. Thus, GloVe is the best technique selected for Symptom2Disease dataset and BERT for Mental Health. Next, the application of the best word embedding technique with the optimal hyperparameters for each classifier resulted in good performance in terms of accuracy, precision, recall, and F1-score. For Symptom2Disease, the best combination is GloVe+SVM with 94.17% accuracy, 94.24% precision, 94.90% recall, and 94.14% F1-score; for the second dataset, it’s BERT+SVM with 92.42% accuracy, precision, recall, and F1-score.</p>

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

WET-ML-GA: A Genetic Algorithm Approach for Optimal Word Embedding and Hyperparameter Tuning in Medical Text Classification

  • 期刊论文

  • 发表日期：

    2025年8月18日

查看全文
  • Meryem Amri,
  • Fatima Zahrae El-Hassani,
  • Khalid Haddouch

摘要

In natural language processing, text classification is a significant challenge, requiring effective word embedding techniques and machine learning models to achieve optimal results. In this paper, we propose a method to enhance machine learning models through the optimal selection of word embedding techniques and model hyperparameters using genetic algorithms. We explore three popular word embedding techniques, namely Word2Vec, GloVe, and BERT, and integrate them with four different machine learning algorithms for text classification. The genetic algorithm is used to simultaneously optimize word embedding techniques and machine learning hyperparameters, identifying the best combination for each classifier to achieve optimal accuracy. The proposed WET-ML-GA approach was tested on two medical datasets, Symptom2Disease and Mental Health. Experimental results demonstrate that genetic algorithms effectively select the optimal configurations, resulting in enhanced performance. In particular, for each dataset, our approach selects the best word embedding technique and the best hyperparameters for each classifier, based on accuracy. Thus, GloVe is the best technique selected for Symptom2Disease dataset and BERT for Mental Health. Next, the application of the best word embedding technique with the optimal hyperparameters for each classifier resulted in good performance in terms of accuracy, precision, recall, and F1-score. For Symptom2Disease, the best combination is GloVe+SVM with 94.17% accuracy, 94.24% precision, 94.90% recall, and 94.14% F1-score; for the second dataset, it’s BERT+SVM with 92.42% accuracy, precision, recall, and F1-score.