Pattern Recognition Neural Network and Class of Grades

Abstract

We investigated the ability of neural network to recognise pattern by using pattern recognition neural network tool box to predict the class of Science laboratory technology students in Federal polytechnic Ilaro.  A total final results of 431 students of the department between 2005 and 2010 was used. The data contained results of ten courses offered by the students together with their GPA and the grade of each student. The marks scored by 420 students in each course together with the grade obtained in each course were used for training, testing and validation of the neural network. 11 student’s data were used for prediction purpose. We used a two-layer feed-forward network, with sigmoid hidden and output neurons (newpr), which can classify vectors arbitrarily well, given enough neurons in its hidden layer. The network was trained with scaled conjugate gradient backpropagation (trainscg). The marks scored by 294(70%) students in each course together with the grade obtained in each course were used for training, 63(15%) for testing and 63(15%) for validation of the neural network. The Grade were classified into five categories: Probation, Pass, lower credit, Upper credit and Distinction. We predict the type of grade based on scores of the students in each course. The training, validation and test were performed with different neuron numbers in the hidden layer i.e 20,15, 10 and 5 neurons. The results showed that among the ANN models, ANN 20 performed best with MSE= 1.511623803293947e-07 and Confusion= 0.061224489795918 followed by ANN 5 with MSE=1.804838395630207e-07 and Confusion= 0.052721088435374. Our research showed clearly that Neural network pattern recognition tools can predict student grade perfectly well if given enough data to train.

Keywords: Pattern recognition, training, validation, neuron, confusion matrix