Performance of Data Mining Algorithms for Classification of Likert Scale Data

The Likert scale enables a systematic assessment of individuals' attitudes and perceptions on specific topics. This scale helps quantify qualitative data by allowing respondents to rate their opinions. Due to their ordinal structure, Likert scales require careful handling during analysis as they influence statistical methods. The work of data mining involves searching through a large repository of data with hopes of finding critical clues that can be used to offer solutions, forecast trends, or identify prospective activities. Data mining also involves establishing relationships, finding correlations to tackle problems, and creating actionable knowledge in the process. This thesis aims to utilize the enormous capabilities of data mining for knowledge discovery in Likert-scale data types and highly imbalanced structures. To compare the classification success of different data mining techniques on Likert-scale data types, the Turkish Family Structure Survey (TFSS) conducted by the Turkish Statistical Institute was selected as the dataset. The experiments were carried out in two stages, firstly, feature selection was performed, and 10 valuable features were determined with the Information Gain criterion. In the classification without imbalance correction, the most successful classification performance was seen in the CART algorithm in the data set with five categories.