An Empirical Approach towards the Knowledge Extraction in a High-Dimensional Clinical Data Set

Abstract

Despite numerous advances in the health care domain, the occurrence of medical errors remains a persistent problem. Human factors such as fatigue, inattention, memory lapse, lack of knowledge and skills, communication failures, use of poorly designed equipment, noisy working conditions, interruptions, and numerous other personal and environmental factors are identified as the major reasons behind the increased medication error rate. Making error is human, but it should be tried to minimize the occurrence of such errors. Thus this has prompted the development of solutions for assisting health care professionals in their decision making process. In this study, we explore the potential of advanced computer technologies for analyzing a time series, high dimensional clinical data set and to what extent they can capture the important medical knowledge hidden in it. The experiments are conducted using two supervised machine learning algorithms namely Artificial Neural Networks and Decision Tree algorithms. As the first experiment, we study the potential of neural networks and decision tree for correctly classifying a given patient data instance into the 3 types of medical conditions considered in the data set. As the second experiment, potencial of a computaional approach for detecting parameter based adverse trends was studied. Validity of the proposed learning models are then experimented using confusion matrix analysis and ROC curve analysis. Results obtained for the exploratory studies carried out in this research, revealed that these kind of computational approaches have a considerable potential for analyzing high dimensional data records and to extract the medical knowledge included within them.