Critical condition classification of patients from ICCDR,B hospital surveillance data

Abstract

During epidemic, when large number of patients appears in short interval, computer models could help in predicting the critical condition of newly admitted patients based on historical information of similar type of patients. In this research, we have developed two classification models by neural network and logistic regression to predict the critical condition of newly admitted patients. Three class labels, i.e., low, medium and high are used in this research to represent the critical condition of patients. However, due to class imbalance problem, the classifier performance was not good for high and mid classes. Therefore, a balancing technique is adopted by using synthetic minority over-sampling technique (SMOTE) algorithm coupled with locally linear embedding (LLE). Experimental results demonstrate that our balanced model outperforms other models by taking care the unbalance nature of ICDDR,B hospital surveillance data.