Cancer Text Article Categorization and Prediction Model Based on Machine Learning Approach

Many research studies have been carried out to identify different variations of cancers in the human body throughout the globe. Due to numerous volumes, the primary concern for healthcare researchers is to segregate the article into a well-thought-out manner. Biomedical text classification is a globally volitional realm due to its numerous uses in different fields. This study motivates us to use a dataset related to cancer articles text containing 7570 research papers with categories into three classes. Like the Word2Vec and BERT models, the word embedding technique converts the text into a numeric vector. The text classification-based prediction model considers nine Machine Learning (ML) classifiers. The models are tested with k-fold cross-validation and evaluate the prediction performances by considering the different standard metrics. The model trade-off between bias and variance is controlled by calculating the standard deviation and k-fold mean accuracy to overcome the overfitting problem of the deployed model. The outcome is observed on the Random Forest (RF) classifier in the case of Word2Vec embedding with a mean accuracy of 99.98%. The BERT models also performed well, with about 97.70% mean accuracy and other parameters compared to the Word2Vec model. The other parameters’ evaluation score was near 1.00, promising in multiclass text classification deployed model performances.