A contemporary feature selection and classification framework for imbalanced biomedical datasets

摘要

Due to the availability of a large number of biomedical documents in the PubMed and Medline repositories, it is difficult to analyze, predict and interpret the document’s information using the traditional document clustering and classification models. Traditional document clustering and classification models were failed to analyze the document sets based on the user’s keyword and MESH terms. Due to the large number of feature sets, conventional models, such as SVM, Neural Networks, Multi-nominal na?ve bayes have been used as feature classification, where additional text filtering measures are typically used as feature selection process. Also, as the size of the document’s increases, it becomes difficult to find the outliers using the document’s features and MESH terms. Biomedical document clustering and classification is one of the essential machine learning models for the knowledge extraction process of the real-time user recommended systems. In this paper, we developed a novel biomedical document feature clustering and classification model as a user recommended system for large document sets using the Hadoop framework. In this model, a novel gene feature clustering with ensemble document classification was implemented on biomedical repositories (PubMed and Medline) using the MapReduce framework. Experimental results show that the proposed model has a high computational cluster quality rate and true positive classification rate compared to traditional document clustering and classification models.