A hybrid feature selection method developed for use in the classification of turkish texts

With the spread and use of technology in every field and the ease of access to the internet, it has become easier to produce, share and store content. As a result of these interactive actions, large increases have been observed in the amount of data stored from past to present. Most of the stored data consists of texts. With the increase in the amount of data, there has been an increase in storage costs. As a result of the significant and irrational increase in data sizes, it has emerged that data should be managed well. In addition, the high number of features in studies using data is one of the factors that increase the cost and decrease the performance. With a good analysis, it is also possible to obtain sector-independent gains from these data. In order to get rid of the cost and performance problems caused by the high dimensionality of the texts and to increase the prediction success rates, it is necessary to choose among the features. At this point, feature selection methods should come into play. In this study, a hybrid feature selection method has been studied for the classification of Turkish texts. The aim of the method used in the study is to create a hybrid feature selection method that measures the quality of the features with the majority of votes and rank allocation among all features in order to combine weak filters such as information gain, symmetric uncertainty and correlation based. Attribute Selection. Obtained features were tested with algorithms such as Naive Bayes, J48, Support Vector Machine, Random Forest and the results were observed. According to the results obtained, more positive results were obtained compared to other filtering methods.