Patients admitted to the ICU suer from critical sickness or injury and are athigh risk of death. So predicting mortality in patients hospitalized in ICU isessential to measure the Severity Of Illness (SOI) and decide about the valueof novel therapy, interventions, and healthcare policies. Moreover, estimatesof mortality risk can be helpful and practical in triage and resource allocation,in determining proper levels of care, and also in discussions with patients andtheir families around predictable issues.As a consequence, the severity of a patients illness can be determined by criticalphysiologic, clinical, and demographic factors. For the purpose of assessingmortality risk, it is necessary to study aggregate data from large, heterogeneousgroups of patients. The remarkable utilizes of mortality prediction in theICU are also in the areas of health research and administration, which ofteninclude looking at groups of critically ill patients.Traditionally, such population-level studies have been extensively approved asapplications of mortality prediction given the cohort-based source of predictionmodels. In this respect, to compare the average severity of illness betweengroups of critically ill patients and between groups of patients admitted inclinical trials, mortality prediction is utilized. Furthermore, for the purposeof benchmarking and performance evaluation of health systems and ICUs,predicted mortality can be compared with observed mortality rates.Several Severity Of Illness (SOI) scores have been introduced in the ICU,which are used to predict outcomes including death or alive. Some of thescores are as follows: Acute Physiology And Chronic Health Evaluation (APACHE) Simplied Acute Physiology Score (SAPS) Mortality Probability Model (MPM) Sequential Organ Failure Assessment (SOFA)The result shows that the aforementioned scoring systems perform well,with Areas Under the Receiver Operator Characteristic Curves (AUROCs)generally between 0.8 and 0.9. In order to improve prediction accuracy, theresearch is investigating methods to leverage the enhanced completeness andexpressivity of modern Electronic Medical Records (EMRs). Specically, the4granular nature of EMRs can lead to creating a personalized predictive modelfor a given patient by identifying and utilizing data from similar patients.2 Study DatasetThe objective of this study is to create mortality prediction models using therst ICU admissions from all adult patients in the Medical Information Martfor Intensive Care (MIMIC)-II version 2.6. A total of 24,581 ICU admissionsin MIMIC-II are taken into consideration.Moreover, following demographic/administrative variables were extracted tobe used as features for the classier: age at ICU admission gender admission type (elective, ur-gent,emergency) rst ICU service type of the ICUadmission heart rate mean and systolic blood pres-sure (invasive and noninvasivemeasurements combined) body temperature Saturation Of Peripheral Oxy-gen (SpO2) respiratory rate creatinine potassium sodium chloride bicarbonate hematocrit white blood cell count glucose magnesium calcium phosphorus lactateThe aforementioned measurements were obtained at the beginning of theICU admission or at most within the rst 24 hours for the great majority ofthe ICU admissions in MIMIC-II. However, the very rst measurements timewith respect to the ICU admission time would have varied between patients.5Mortality at 30 days post-discharge from the hospital was extracted to predictthe patient outcome.