Early Admission Selection Process Into Sixth Form Science Streams Using Neural Networks Model

• Main Author: Wong, Tuck Sung
• Format: Thesis
• Language: English; English
• Published: 2000
• Subjects: QA76 Computer software
• Online Access: https://etd.uum.edu.my/226/1/WONG_TUCK_SUNG_-_Early_admission_selection_process_into_sixth_form_science_streams_using_neural_networks_model.pdf; https://etd.uum.edu.my/226/2/1.WONG_TUCK_SUNG_-_Early_admission_selection_process_into_sixth_form_science_streams_using_neural_networks_model.pdf; https://etd.uum.edu.my/226/

The purpose of this study is to investigate and offer a model, based on Neural Networks Theory, capable of selecting successful students for early admission into sixth form science streams. This model would be able to perform the intended selection process even before the results of the Sijil Pelajaran Malaysia (SPM) are announced. The main benefit of this early admission was to allow students to start their classes early and to complete their demanding syllabus on time. The noble motive was to save time and to prevent time wasting, which would be true if students had to wait until the examination results are announced before they could start their classes. A neural networks solution, using Multi Layer Perceptron (MLP) and Steepest Gradient Descent algorithm, was studied to offer a better model to select students more meticulously. A total of 1488 data samples from ten secondary schools in the silver state of Perak, consisting of past-year Form 4 and Form 5 internal examination results, were collected in order to be trained and tested using the Neural Connection Version 2 software. A correct prediction of 92.18% accuracy was achieved using this Neural Networks model. Analysis of the data showed a reasonably strong correlation between the input variables, which consisted of subjects’ marks, aggregates and grades achieved, with the targeted output variable, which was the offer to continue with Sixth Form. It also showed that the data were only slightly skewed and were normally distributed. The trained Neural Networks model was found to produce a comparable accuracy when applied to other data from either only boys or girls schools, and from either urban or rural schools.