Using and comparing different decision tree classification techniques for mining ICDDR,B Hospital Surveillance data

In this research we have used decision tree induction algorithm on Hospital Surveillance data to classify admitted patients according to their critical condition. Three class labels, low, medium and high, are used to distinguish the criticality of the admitted patients. Several decision tree models are developed, evaluated, and compared with different performance metrics. Finally an efficient classifier is developed to classify records and make decision/predictions on some input parameters. The models developed in this research could be helpful during epidemic when huge number of patients arrive daily. Due to rush of duty doctors and scarcity of required number of physicians, it is hard to diagnose every patient. Any computer application could be helpful to diagnose and measure the criticality of the newly arrived patient with the help of the historical data kept in the surveillance database. The application would ask few questions on physical condition and on history of disease of the patient and accordingly determines the critical condition of the patient as low, medium or high.

► We use decision tree models to classify patients according to their criticality.
► Various data preprocessing and modeling activities are carried out to reduce features from data set.
► Different decision tree models are generated by considering equal and unequal instance distribution from classes.
► Performance improves if we consider equal distribution of instances from each class.