Data Extraction for Associative Classification using Mined Rules in Pediatric Intensive Care Data

Abstract

Based on the characteristics of health and medical informatics, data mining techniques that were designed to tackle healthcare problems are faced with new challenges. One such challenge is to prepare medical data for pattern mining or machine learning. In this paper, we present a feature engineering technique for the Associative Classification of the Systemic Inflammatory Response Syndrome (SIRS) in severely ailing children by mining Associative Rules. SIRS is characterized as the body's excessive defense response due to malevolent stressors such as trauma, acute inflammation, infection, malignancy, and surgery. It can have an impact on the clinical outcome and elevate vulnerability for organ dysfunctions. We aim to extract the features from given datasets using a specific extraction process and after the transformation, those features are used to mine rules using Association Rule Mining. Those rules are used to perform Associative Classification and evaluated with the result generated by SIRS criteria defined by the experienced clinicians. The mined rules provide better control over sensitivity and specificity than the SIRS criteria.