Development of Data Mining Software Framework by Using Map/Reduce
Method ın Cloud Computing Systems Machine learning allows specially solution of classification and regression problems. Support vector machine algorithm (SVM) is the most commonly used classification method among machine learning techniques due to its high generalization property. However, SVM needs high computational requirements for high-dimensional datasets.
In this study, multi-class support vector machine algorithm over cloud computing systems with MapReduce technique is studied. This work can be divided into four parts.
In the first section, general information on cloud computing systems is provided. Service models, distribution models, cloud computing systems for scientific research, functional programming and MapReduce in cloud computing were examined.
In the second part, SVM algorithm is analyzed. The use of SVM algorithm in classification and regression is studied.
In the third part, SVM classification algorithm with MapReduce technique of training high-dimensional datasets on a distributed cloud computing system servers are described. The historical development of MapReduce technique and functional programming that commonly used in cloud computing systems is described.
The fourth chapter of this thesis is the application part. It consists of two sections. In the first section, using text and digit classification datasets that is provided by University of Caroline Irvine (UCI) for machine learning, SVM classification algorithm that allows only binary classification is used for multi-class classification with some techniques. In the second section, social media posts data set of foundations and state universities in Turkey is classified. The models that are created with MapReduce are tested with 10-fold cross-validation technique and accuracy improvement of each iteration is shown with graphics.
