An Expansive Performance Analysis and Comparison between Different Supervised and Unsupervised ML Algorithms for Categorization of ICU Patients at an Indian Hospital
Amrita Ghosh, Partha Pratim Das, Subrata Jana, Bibhas C. Giri, Anirban Sarkar
Abstract
Purpose: Accurate predictive decisions and analysis on actual data can be a fruitful capability, opportunity and cost effective solution in healthcare practice. After ICU transfer Patient's situation depicts the status of hospital's post ICU care. Research Objective: This study emphasizes the evaluation and comparison of the performance of data mining methods on a hospital ICU database. This paper aims to build ML models that predict patient's outcomes in terms of three categories: Continuing in ICU, Discharge/shifting in General bed, Mortality cases. Methods: This categorized model performs on the multi parameters of 18 or above aged ICU patients' database, collected from Indian hospitals. Here, Random Forest (RF), AdaBoost, Bagging, Logistic Regression, Decision Tree, KNN are deployed as Supervised Learning Algorithms. For Unsupervised Learning Algorithm K-Means Clustering and Gaussian Mixture Model are used. 30% data are tested and 70% data are trained as whole. Results and Analysis: Proposed model shows the accuracy score, AUROC-curve, performance and confusion matrices for both Supervised and Unsupervised Learning Algorithms. Conclusion: This model emphasizes the importance of patients’ outcome after ICU transfer in medical and data science sector though several scoring and data mining systems are available.