Abstract: MATH/CHEM/COMP 2002, Dubrovnik, June 24-29, 2002
MACHINE LEARNING ANALYSIS OF CALCIUM, OXALATE AND CITRATE INTERACTION IN IDIOPATHIC CALCIUM UROLITHIASIS IN CHILDREN
Danko Milosevic1, Danica Batinic1, Paško Konjevoda1, Nenad Blau2,
Nikola stambuk3, Ana Votava-Raic1, Ljiljana Nizic1, Kristina Vrljicak1, and Danko Batinic1
1Children's University Hospital, Salata 4, HR-10000 Zagreb, Croatia
2Children's University Hospital, Zurich, Switzerland
3Rudjer Boskovic Institute, POB 180, HR-10002 Zagreb, Croatia
The role of urine oxalate, calcium and citrate interaction in idiopathic calcium oxalate urolithiasis was analysed by means of machine learning algorithms. We examined 30 children with idiopathic urolithiasis and compared them with a group of 15 sex- and age-matched healthy children. OneR and J4.8 classificators, parts of the larger data mining software based on machine learning algorithms for supervised and unsupervised learning and writen in Java programming language - The Waikato Environment for Knowledge Analysis (Weka, version 3.3) were used for discrimination between healthy children and children with urolithiasis. Using OneR classificator, we were unable to induce acceptable classificator for discrimination between healthy children and children with urolithiasis. Contrary, J4.8 classificator was able to discriminate between them. The accuracy of classification with induced decision tree was 97.7% (91.1% with leave-one-out cross-validation technique). Decision tree, constructed with J4.8, pointed out the value of oxalate and citrate regarding calcium. The algorithm analysis shows that complexe interaction between urine oxalate, calcium and citrate as the major promoters and inhibitor of crystallization is the way to estimate their role in the risk of development of urolithiasis.