Prediction of students' performance based on heterogeneous and ambiguous data

Introduction/purpose: The presented research describes the application of artificial neural networks for prediction of students' performance based on heterogeneous and ambiguous data. Methods: The students' data collected during three school years on subject "Object-Oriented Design" at the School of Electrical and Computer Engineering, Academy of Technical and Art Applied Studies in Belgrade represent the basis of the research. The sample size is one hundred fifty-nine students. The initial dataset consists of thirty-six different features from three different sources of collected data. The presented data and results are completely anonymized. The research describes the complexity of data preparation, as well as the processes of modeling, training, testing and evaluation of an artificial neural network, applied for the prediction of students' performance. Results: The main results of the research are the handling the ambiguity and heterogeneity of data and the design of the prediction model. In addition, the research includes discussions about data preprocessing and machine learning techniques applied on prediction models. Conclusion: The development methods, prediction models and prediction results, their critical review, the limitations of the research, and the possibilities for improvements represent the main conclusions of the research.