Classification of family structure research data by data mining algorithms

Similar to the term mining, data mining is the process of discovering useful intelligence, analyzing enormous amounts of information and datasets to help solve problems, predict trends, mitigate risks, and find new opportunities. At the same time, data mining involves building relationships, finding correlations to deal with problems, and generating actionable insights in the process. In this thesis, it is aimed to discover information in Likert scale data types by taking advantage of the enormous capabilities of data mining. To compare the classification success of different data mining techniques on Likert scale data types, Turkey Family Structure Survey (TAYA) conducted by the Turkish Statistical Institute (TURKSTAT) was chosen as the data set. In the experiments carried out in two stages, first feature selection was made, and 10 valuable features were determined with the Information Gain criterion. In the classification phase, firstly, the number of categories in the dataset was changed and the classification success of the algorithms was measured. Then, the imbalance between the classes on the dataset, which is imbalanced due to its structure, was removed and its effect on the classification analysis was observed. The most successful classification performance was observed in the CART algorithm for the dataset with five categories and in the RepTree algorithm for the dataset with three categories. To eliminate the imbalance between classes, three different data sets were created by changing the total sample volume with resampling and data completion method. It was observed that the algorithm with the highest classification success in the created data sets was the CART algorithm.