Using of support vector machines in medical research

In today's technology, with increase of capacity and speed of computers, in various sectors, capacity and complexity of the stored data also increases. As a result of processing of these data with appropriate advanced techniques, the dizzying technological changes have been observed in living conditions. For the purpose of extracting meaningful information from huge amounts of data, data mining methods that found intensive usage especially in recent years and in the future will replace with the traditional statistical methods are preferred. Support vector machine (SVM) is one of these methods and today it is used for classification, estimation or pattern recognition mostly in engineering applications. In this study it was aimed to define concepts of data mining, explain theoretical foundations of SVM particularly and use of this method for diagnosing in medical research. For this purpose, firstly data mining and data mining methods were mentioned briefly, then support vector machines method was described comprehensively. SVM, which is one of data mining methods is a machine learning tool that uses supervised learning to classify or predict the data. Basic idea behind SVM is to classify the data by dividing them with a plane or hyperplane. DVM achives this procedure by doing maximum the margin between two classes. After training of the data, SVM aims to classify the new data correctly. In medicine, SVM is used for cancer morphology, identifying success of treatment and related gene, diagnosing various diseases. In application of the study, informations about 433 patient who were refer to the outpatient department of Zonguldak Karaelmas University Faculty of Medicine between date of 1-31 January 2011 for complaints of night eating syndrome were used. Using these data, variables that is effective in diagnosing were examined with univariate analysis, logistic regression and SVM methods and only GYA_puan variable found to be effective in the three approaches. Also when the classification performances of logistic regression and SVM were examined, it was seen that both of the methods gave similar results but superiority of SVM was discussed. In addition, performances of linear, polynomial, sigmoid and radial basis function (RBF) mentioned most common in applications were compared. In these comparisons, it was seen that the classification performances of 4 different kernel function gave similar results, however superiority of RBF in various ways was mentioned