Applications of Rule Based Classification Techniques for Thoracic Surgery

It has been aroused the necessity of extracting meaningful information from huge amount of available data that is accumulated as result of development in computer technology and database software. Traditional methods can’t cope with turning the data to the knowledge due to amount and complexity of accumulated data that has so many hidden patterns in it. Thus, nowadays the data mining techniques are commonly used for analyzing huge amount of information. Classification, clustering and associated rule extraction of data mining techniques are preferred widely. Classification is the operation of determining class of the data by forming a model that makes use of data whose categories are previously determined. Data mining techniques are frequently used to form a classifier that determines belonging class of a new data among the predetermined classes. Although these classification methods including different classification and rule extraction algorithms are generally successful they don’t reach the required success levels when it comes to multi-class real world problems. In this research the thoracic surgery rules have been induced by classifying thoracic surgery into different classes. For this classification task the rule based methods of Conjunctive Rule, Decision Table, Decision Table/Naive Bayes (DTNB), Java Repeated Incremental Pruning (JRip), Navajo Nation Gaming Enterprise (NNge), One Rule (OneR), Partial C4.5 (PART), RIpple-DOwn Rule learner (Ridor) and 0-R classifier (ZeroR) were used. Correctly Classified Instances were found as % 85.1064, % 84.8936, % 84.8936, % 84.6809, %84.4681, % 83.4043, % 81.7021, % 81.0638 and % 79.1489. These results have shown that ZeroR has the most successful prediction ratio among the four techniques regarding to classification rules. The data used for this study has been published in 13 November 2013. Thus, there is no way of making any comparison with the previous studies. Due to this situation only the classification rules methods used in this study were compared to each other. The quality of rules produced by the methods of this study can be enhanced by using different rule pruning methodologies as future study.