Using neural networks to predict MBA student success

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

Results

Table 1 shows that the neural network can classify students into successful and marginal groups fairly accurately. Table 1 gives the results indicating that the neural network model classified 93.38 percent of the successful and 80.9 percent of the marginal students correctly. The overall prediction accuracy for the neural net work model is 89.13 percent.

To assess the neural network model's ability to classify students into successful and marginal groups, the predictive ability of the neural network model was compared with two statistical models, namely, logistic regression (logit) and probit. O'Leary (1987) recommends validating artificial intelligence models against other statistical models.

Logit and probit are models designed to overcome the problems with a dichotomous dependent variable. The fitted (predicted) values from logit and probit models can be used, like multiple discriminant analysis, to classify entries into mutually exclusive groups using a set of predictor variables. The results obtained using logit and probit models are presented in Tables 2 and 3 respectively.

Table 2 shows the classification matrix obtained from Logit. The sample results indicate that the nine-variable logit model classifies 86.78 percent of the successful and 46.03 percent of the marginal cases correctly. The overall rate of prediction accuracy for the Logit model is 72.83 percent.

Table 3 lists the classification matrix obtained from Probit. The sample results indicate that the nine-variable probit model classifies 87.6 percent of the successful and 46.03 percent of the marginal cases correctly. The overall rate of prediction accuracy for the Probit model is 73.37 percent. Thus, a comparison of the neural network results with those of the two statistical models indicates that the neural network model performs as well as the statistical models.

Summary

This paper has outlined the features of a neural network technique to evaluate and predict MBA student performance in their graduate programs. This artificial intelligence-based technique holds the promise as an evaluation tool to classify MBA students into successful and marginal categories. The technique can be a valuable decision tool to assist university admission officials in identifying students that are likely to be successful in an MBA program. The use of a non-linear model such as the neural network allows administrators to specifically incorporate uncertainties into the decision making process. The traditional check list and formula approaches do not permit this flexibility.

The predictions of these neural network and statistical models are reported in tables 1, 2, and 3. The results show that the neural network model performs as well as the traditional statistical models and is a useful tool to predict MBA student performance. The overall results suggest that the classification accuracy (and implied predictive power) of the neural network model is 89.13 and that of the logit and probit models are 72.83 percent and 73.37 percent respectively. The use of a neural network model can support and potentially improve decision making by MBA directors and deans.