Academic journal article Kuram ve Uygulamada Egitim Bilimleri

Identifying the Classification Performances of Educational Data Mining Methods: A Case Study for TIMSS

Academic journal article Kuram ve Uygulamada Egitim Bilimleri

Identifying the Classification Performances of Educational Data Mining Methods: A Case Study for TIMSS

Article excerpt

(ProQuest: ... denotes formulae omitted.)

Data mining is a multidisciplinary study area that includes many different statistical procedures. The main goal of data mining is to explore useful hidden patterns among huge data sets. Unlike traditional statistical methods such as linear regression, data mining methods do not require the assumptions of linearity, variance, homogeneity, or normality (Sinharay, 2016).

With the rapid development of information technologies, a great number of techniques within data mining have been applied over many disciplines, including social sciences, physics, engineering, and medicine. Studies that use data mining methods in the field of education are generally known as educational data mining (EDM) studies, of which an extensive literature review was performed by Romero and Ventura (2007) for the period between 1995 and 2005. In their study, EDM is stated as an iterative cycle that includes hypothesis formation, testing, and refinement. They pointed out that the outputs obtained from EDM methods guide educators in their discovery of useful information on formative evaluation. Through the usefulness of this newly discovered information, educators have established a pedagogical basis for decisions when designing or modifying an environment or teaching approach (Romero & Ventura, 2007). Another important study by Baker and Yacef (2009) dealt with EDM and its major trends. Their study discussed four different domains within the field of EDM, addressing studies with regard to each of these domains. In addition to these substantial works, Pena-Ayala's (2014) study provided an EDM survey from the beginning of 2010 until the first quarter of 2013 that included 240 published papers. Despite the important studies that exist in the literature on EDM mining, it still lacks research, particularly on supervised learning methods for huge data sets that require computationally difficult algorithms (Sinharay, 2016). One of the biggest known challenges facing education planners is how to analyze huge data sets in terms of student's characteristics such as knowledge, motivation, and attitudes (Baker, 2010). In the process of improving the quality of managerial decisions for future education strategies, understanding and exploring the hidden patterns from observable data is generally difficult and time consuming when done manually (Mohamad & Tasir, 2013). Therefore, much attention should be given to the study of EMD in order to enlighten educators and education planners. Output obtained from EDM can offer need-oriented solutions through different perspectives in the process of determining useful education strategies (Bilen, Hotaman, Aşkın, & Büyüklü, 2014).

Several studies that fall within the concept of EDM have been carried out in the recent literature in order to provide reliable solutions regarding educational phenomena. He's (2013) study pointed out that assessing learning performance, providing feedback, and adapting learning materials based on students' learning behaviors are some of the reasons to use EDM. Cortez and Silva's (2008) study applied four supervised learning methods (i.e., decision tree, random forest, neural networks, support vector machines) for building a model based on the student performance of secondary schools in Portugal. Their findings showed not only the best prediction model but also supported the idea that academic achievement and past performance highly correlate with each other. Ramaswami and Bhaskaran (2010) collected data from higher education students and constructed a prediction model based on CHAID, which is the most commonly used decision tree algorithm. Furthermore, their results identified statistically significant factors influencing academic performance. The studies of Alivernini (2013), Abad and Lopez (2017), and İdil, Narlı, and Aksoy (2016) also dealt with decision trees for constructing student-based models that would ensure accurate classifications and predictions, but respectively used the different algorithms of CART, J48 and C5. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.