Publication: Makine öğrenmesi yöntemleri ile vakıf üniversiteleri doluluk tahminlemesi
Abstract
Makine Öğrenmesi, Sınıflandırma, Destek Vektör Makinesi, k-En Yakın Komşu, Vakıf Üniversitesi ÖZET Bu çalışmada makine öğrenmesinin sınıflandırma yöntemlerinden destek vektör makineleri ve k-en yakın komşu algoritması incelenmiştir. Bu yöntemler kullanılarak vakıf üniversitelerinin izleyen yıl ya da yıllarda kontenjanlarını doldurup doldurmayacağının tahmin başarısı araştırılmıştır. Araştırmada kullanılan veri seti, Ekim 2012 – Şubat 2013 tarihleri arasında İstanbul ilinde faaliyet gösteren ve fakültelerinde işletme bölümü bulunan vakıf üniversitelerini kapsamaktadır. Bu kapsamda 22 vakıf üniversitesine ait 32 öznitelik kullanılmıştır. Veri seti, öznitelik seçimi öncesi ve sonrası olmak üzere ayrılmıştır. Bu seçim asgari gereksizlik-azami ilişkisellik (mRMR) yöntemi kullanılarak yapılmıştır. Sınıflandırma performansları birini dışarıda bırakma ve k-katlı çapraz geçerleme yöntemleriyle değerlendirilmiş, destek vektör makineleri ile elde edilen sonuçlar k-en yakın komşu algoritması ile karşılaştırılmıştır. Bu çalışmada, destek vektör makineleri ve k-en yakın komşu algoritmasının kontenjan doluluğunun tahmin edilmesinde tutarlı ve tahmin gücü yüksek, tatmin edici sonuçlar verdiği görülmüştür. FOUNDATION UNIVERSITIES OCCUPANCY ESTIMATION THROUGH MACHINE LEARNING METHODS Machine Learning,Classification, Support Vector Machine, k- Nearest Neighbor, Foundation University
In this study, support vector machine and algorithm of the k-nearest neighbor, which are methods of classification process of the machine learning, have been studied. By using these methods, the success of the filling quotas for the foundation universities has been searched for the future years. The data set used in the research includes the İstanbul located foundation universities which have business administration departments in their faculties between the dates October 2012 and February 2013. The study focuses on 32 features which belong to 22 foundation universities. The data set has been divided into two categories as pre and post feature selection. This selection has been applied by the minimum-redundancy maximum-relevance (mRMR) method. Classification performances have been evaluated by the methods of leave one-out and K-fold cross validation,the results reached with the support vector machines have been compared with the algorithm of k-nearest neighbor. The results of this study have revealed that support vector machine and algorithm of the k-nearest neighbor methods are consistent and successful in giving high prediction rate in terms of filling quotas of the foundation universities.
In this study, support vector machine and algorithm of the k-nearest neighbor, which are methods of classification process of the machine learning, have been studied. By using these methods, the success of the filling quotas for the foundation universities has been searched for the future years. The data set used in the research includes the İstanbul located foundation universities which have business administration departments in their faculties between the dates October 2012 and February 2013. The study focuses on 32 features which belong to 22 foundation universities. The data set has been divided into two categories as pre and post feature selection. This selection has been applied by the minimum-redundancy maximum-relevance (mRMR) method. Classification performances have been evaluated by the methods of leave one-out and K-fold cross validation,the results reached with the support vector machines have been compared with the algorithm of k-nearest neighbor. The results of this study have revealed that support vector machine and algorithm of the k-nearest neighbor methods are consistent and successful in giving high prediction rate in terms of filling quotas of the foundation universities.