Abstract:

Analysing and modeling the performance of the students can facilitate educational systems to improve their quality and help students to perform better in their studies and therefore have a better future!! Bayesian networks can model the causal interactions and statistical relationships of a system’s variables with a graphical demonstration which is easy to interpret. Using such models and having evidence about one or many of performance indicators of students, it is possible to investigate the status of other indicators in the model. It is also possible to predict the effect intervention on one or more indicators on the other parts of the network. Development of machine learning techniques for Bayesian networks in the recent years makes it possible to discover the knowledge of a domain automatically using the collected data.

The data is collected based on previous research conducted by Niemivirta (2012). The collected data measures students academic achievements and goal orientation. In this study, the Bayesian networks is used to model the causal relation of student’s performance factors and then the model is used to classify the students according to performance and explore the effect of the intervention on each individual students. We elaborate how degree program may anticipate drop out by affecting to some key dropout factors such as fear of failure and academic withdrawal.

All the real-life datasets have missing values to some extent due to failure in the recording, human errors in sensors or a non-response in a survey. There are four types of missing values: Missing completely in Random (MCAR), Missing at Random (MAR), Missing Not at Random (MNAR), and Not Missing at Random (NMAR) which are dependent on an unobserved variable. The last types of missing values can be called Filtered values which are not missing in reality. These are the values, which their existence is depending on the values of other variables. (Conrady and Jouffe 2007). One approach to handle the missing values, especially when dealing large amounts of data, is to remove the records with the missing data. Koller and Friedman (2009) had shown that this approach can potentially change the distribution of the variables and lead to a magnificent amount of bias in the dataset. Finding the best network in the search space of all possible network is NP-hard, and the heuristic search algorithms can easily trap in a local minima. To tackle this problem, Munteanu and Bendou (2001) developed the EQ framework to use the space of Essential graphs of an Equivalent class of BNs to search for a suitable graph.

To achieve the study objectives we aim to answer the following research questions.
Are students with strong goal orientations less likely to drop out?
Do students with different socio-geographical characteristics have different patterns of motivational beliefs?
What approach HAAGA-HELIA UAS needs to take to anticipate dropouts?

The data gathered by quantitative approach is analyzed using parametric tests. The distribution of parametric tests is powerful than nonparametric test, however, the nonparametric test is much more flexible. (J. P. Verma 2013) The data gathered by two methods also defined as qualitative data and quantitative data. Quantitative research quantifies opinions, behaviours and other defined variables.

Keywords:

Bayesian Analysis, Data mining, Performace prediction.