Clinical data driven framework For risk assessment and medical Diagnosis recommendations in Intensive care unit

Abstract

With the proliferation of advanced technologies in healthcare sector, there is newlinea growth of medical data, which plays an important role towards medical research newlineand development along with patient database management. Such form of medical newlinedata is quite massive in size and complex in its structure of retention. This newlineeventually causes a potential impediment to understand the degree of severity of newlinemortality factor associated with the diagnosis of critical disease of the patient when newlineadmitted to the hospital. Therefore, the proposed study considers a standard MIMICIII newlinedataset for this purpose which has enriched data associated with the critical newlinedisease for a patient during his ICU stay. An exhaustive review of existing studies is newlinecarried out toward analyzing medical data to find out that MIMIC-III dataset has not newlinereceived much attention in comparison with other schemes of medical data analysis. newlineFurther, it is noticed that learning-based schemes are predominantly used in newlinepredictive analysis of medical data; however, they are highly inclined towards newlinehigher accuracy at the cost of computational complexity. Therefore, the proposed newlinestudy introduces a novel learning-based scheme in order to predict severity of risk newlineassociated with the critical disease complication and rate of mortality of patient newlineadmitted in ICU. The complete implementation is classified into two modules i.e., i) newlinecomputational framework for feature engineering and ii) integrated learning newlinetechnique for predictive diagnosis. The first module is meant for enriching the newlinequality of data by performing unique preprocessing operation. According to the first newlinemodule, the attributes associated with diagnosis and procedures are extracted from newlinemetadata structure of predictors from MIMIC-III dataset, which is followed by up newlineseries of feature engineering operation resulting in quality data. This model is further newlineassessed with multiple machine learning modules to find that ANN offers better newlinecompatible and supportive analytical performance in comparison with KNN and newlineLogistic