Mining Students' Performance in UPSR Using Statistics and Neural Networks

Academic performance has become an important evidence of determining the quality in Malaysia's education system. The examination data is collected on the previous students' examinations yet to be tested for their coming UPSR. The other related data such as family background and schooling information are also involved. The raw data is preprocessed and analyzed using statistical method. The results from the statistical analysis indicate the significant contribution of these attributes to the achievement model. The combinations of input variables, hidden layer and output nodes are explored to predict the students' performance. Five models are constructed based on five subjects to relate them with other factors for the purpose of descriptive analysis. The relationship between examination results and other factors are investigated thoroughly to
enhance the prediction model. The performance model obtained in this study uses parameters such as; learning rate 0.1, momentum rate 0.1, Sigmoid activation function,
100 epoch learning stopping criteria with its architecture, 13 inputs unit, 2 hidden units and 5 output units. The result indicates that Neural Networks has high potential to be used in predicting students' performance.