Destek vektör makineleri ile tiroid hastalıkları tanısı

The thyroid gland is the organ that is located on the anterior side of our neck. The duty of thyroid gland is to produce and stock thyroid hormone and to regulate the metabolism by transferring thyroid hormone to the blood when it is needed. When thyroid gland does not secrete enough amount of hormones, pituitary gland makes thyroid gland produce more hormone by increasing TSH secretion. Shortage of the hormones of thyroid gland is called hypothyroid. If thyroid gland secretes too much hormones, TSH hormone, secreted from pituitary gland, decreases. That is, the more T3 and T4 hormones are in our blood, the less is the TSH hormone. This condition is called hyperthyroid. All cells in our body are affected by thyroid hormones. The growth of human in mother's womb, after birth and all metabolism functions are controlled by thyroid hormones. There is almost no organ or cell that is not affected by thyroid hormones. Therefore early diagnosis of thyroid disease is undoubtedly important. Support Vector Machine (SVM) is an important learning machine that is based on a search for an optimal separating hyperplane that is able to separate the samples of two different classes. This research aims to construct a system based on classification via SVM for diagnosis of thyroid diseases. Since it is the matter of decision, the diagnosis of diseases can be predicted by classifiers. Having been a quite popular classification algorithm, SVM is among the best classifiers to deal with this duty. Training and testing data consists of test results of 215 different people taken from a machine learning repository. 35 samples of hyperthyroid patients , 30 samples of hypothyroid patients and 150 healthy samples are used. Input space of data consists of 5 different inputs (T3RU, T4, T3, TSH and MAD-TSH). Feature subset selection is used to increase the accuracy of the corresponding classifier. By using feature selection, one can easily decrease the number of features. In this thesis Fisher Score Algorithm is used to perform this preprocessing procedure. The most important features in classification are obtained by Fisher Score Algorithm. Then the most important features are used to train the network. Another important feature in an SVM classifier is the selection of parameters values used in classification. Parameters are what make a classifier manipulatable in order to successfully classify the patterns. One of the most important parameters in SVM is slack variable. Slack variable is used when the data is not linearly or non-linearly separable (this could happen when there is noise or when it is really impossible to separate the patterns). Using slack variable prevents SVM to create useless optimally separating hyperplanes. The weight of slack variable is adjusted by a coefficient called Soft Margin constant "C". Normally parameter C is adjusted by the user considering the characteristics of dataset. However, it is not easy to decide which value for C is optimal. This problem is handled by one of the most popular optimization algorithms called "Genetic Algorithm". Genetic Algorithm is an optimization algorithm that is inspired by the nature of evolutionary process. In this thesis, an expert system is developed for the diagnosis of thyroid diseases, by combining these three methods.