Significance of Healthcare Analysis Using Word Embedding With Machine learning Classifiers

Abstract

Abstract The forecasting of health news is becoming the main issue to divide healthcare news into different classes based on user comments. Moreover, predicting class content is significantly important in influencing news forecast reports. In this paper, the news data from the MIND dataset considered classes lifestyle and health with user comments, in the preprocessing stage tokenization, stop word removal, stemming, lemmatization etc. are performed. After the preprocessing step obtained words need to be converted into numerical vectors and these vectors along with their weighted values are determined for Modeling, in this way applied to some techniques like Word embedding, feature selection, SMOTE (synthetic minority class over-sampling technique) and various ML (machine learning) classifiers To find the impact of used techniques for the given input data and the performance of all used techniques is obtained by some descriptive statistics analysis and Rank test. Finally, the comparative analysis is shown by the performance of accuracy and F-measure parameters for each technique. Experiments show that the LOGR (logistic regression) and a variant of SVM outperform all other classifiers and it is seen that in this work GLOVE variant is a better embedding technique where CCRA (cross correlation analysis) and SMOTE are better feature selection and class imbalance techniques and their combined effects helps to achieve significant improvement for the models to predict the classes priority based on news comments by users.