Using neural networks to predict MBA student success

College Student Journal, March, 2004 by Bijayananda Naik, Srinivasan Ragothaman

Data Description

Files of one hundred and eighty four recent graduates of a medium-sized, AACSB accredited mid-western University were examined. The graduates in this sample were predominantly of Caucasian race. Sixty-four students were enrolled in the on-campus program and 120 students were enrolled in the off-campus program. Courses were offered in the evening at two off-campus locations. We classified students with a graduate GPA of 3.3 or more as "successful" and students with a graduate GPA of less and 3.3 as "marginal" and used a dichotomous (1 = successful, 0 = marginal) variable as the dependent variable (see Hardgrave et. al., 1994).

In this research, ten explanatory variables were used, namely, campus location (LOCATION), citizenship status (CITIZEN), gender (SEX), ethnic status (RACE), 4-year undergraduate grade point average (4UGPA), junior-senior year grade point average (2UGPA), graduate management admission test score (GMAT), age of the student (AGE), undergraduate institution (COLLEGE), and undergraduate major (MAJOR). Undergraduate institution (COLLEGE) variable was coded using the information provided in the Peterson's National College Data Bank. Undergraduate major was either business/economics (one) or other (zero).

Neural Networks

With the advent of modern computer technology and information science, sophisticated information systems can be built that can make decisions or predictions based on information contained in available past data. Such systems are called learning systems and are currently used for the purpose of classification and prediction (Weiss and Kulikowski, 1991). Most of the early applications of neural networks have been in systems such as signal stabilizer, word recognizer, process monitor, sonar classifier, and risk analyzer etc. Other applications of neural networks found in literature include the fields of aerospace, electronics, telecommunications, manufacturing, medical, banking, securities market, speech recognition, oil and gas exploration, and market research. In this paper, we have attempted to use the neural networks technique in the field of education.

A neural network consists of a network of neurons. Each neuron is associated with an input vector, a weight vector corresponding to the input vector, a scalar bias, a transfer function and an output vector (Demuth and Beale, 1995). A neural network may consist of one or more neurons in each layer. In a network, the final layer is called the output layer and all previous layers are called hidden layers. In the hidden layers, the output of a layer becomes the input for the following layer. The transfer function of a neuron converts the input to the output of the neuron. Multilayer neural networks are quite powerful tools used in solving many different problems. Different types of neural networks are available for different purposes. In this research, a multilayer back propagation neural network is used.

Building The Model

The first step in using neural networks is to design and test alternative networks. The network that is selected for use should have the lowest error rate. Normally, more number of layers and more neurons in each layer can produce a better network with lower error rate. However, there is a real danger of over-fitting if too many layers and/or neurons in each layer are used. Over-fitted models give high error rate and do not classify or predict accurately. Thus, designing a neural network is a trial and error process. In this research, a neural network with two hidden layers and one output layer (total three layers) was used. The output layer has only one neuron and one output, namely, the type of student, successful or marginal. The neural network model was built using a neural network tool available with the MATLAB software package (Demuth and Beale, 1995). The models were trained and tested with 184 data points described in the data description section.