Analysis on the prediction of central line-associated bloodstream infections (CLABSI) using deep neural network classification

Abstract The surveillance dataset of central line-associated bloodstream infection (CLABSI) in agreement with healthcare-associated infections (HAIs) is treated as a significant measure for the distribution of problems associated with CLABSI. The data, collected through several modes, acts as a basis for core statistics in preventing CLABSIs, using data mining. The validation of HAI data is considered a key element in ensuring improved data quality between the users. The increased CLABSI data leads to uncertain information and this affects the accuracy in predicting the comparisons between hospitals, affecting their reputations. The past studies on prediction of CLABSI report only the manual clinical observations, which are not accurate due to the redundancy of data, and they are time consuming. Recently, a few machine learning approaches have been modeled to predict the class of CLABSI; however, the data dimensionality issues have not been addressed. These machine learning models further lack a proper modeling framework that could resolve the problems of classification. In this chapter, we provide a state-of-the-art deep learning prediction model for CLABSI. A deep learning classifier, namely the deep neural network classifier (DNN) algorithm, is used for training using labeled data to classify the variables or features. The accuracy of the classifier is fine-tuned using a sparse minimax concave ridge support vector machine (SMCR-SVM). The trained labeled data is used for diagnosing the test cases to accurately detect the class of CLABSI collected across various datasets. The simulation results are estimated in terms of various performance metrics that include accuracy, sensitivity, specificity, mean absolute percentage error (MAPE), F-measure, precision, and geometric mean (G-mean). The results show the proposed DNN-SMCR-SVM classifier achieves a higher rate of classification accuracy than the other existing classifiers (extreme gradient boosting, logistic regression, supervised machine learning, unsupervised machine learning, and ensemble learning).