The following information was submitted:
Transactions: INTERNATIONAL JOURNAL of APPLIED MATHEMATICS AND INFORMATICS
Transactions ID Number: 19-379
Full Name: Hu Hongping
Position: Associate Professor
Age: ON
Sex: Female
Address: College of Science,North University of China,Taiyuan, Shanxi 030051
Country: CHINA
Tel:
Tel prefix:
Fax:
E-mail address: huhongping@nuc.edu.cn
Other E-mails: hhp92@163.com
Title of the Paper: classification of the students¡¯ scores based on some artificial neural networks
Authors as they appear in the Paper: Hu Hongping Bai Yanping
Email addresses of all the authors: huhongping@nuc.edu.cn,baiyp@nuc.edu.cn
Number of paper pages: 8
Abstract: In this paper, the data of students¡¯ scores are analyzed by using the nonlinear BP neural network algorithm with a hidden layer, the probabilistic neural network algorithm, the perceptron algorithm and self-organizing compete neural network algorithm. We take 3000 students¡¯ scores on only course to be analyzed and classified. Among these scores, 121 students¡¯ scores are trained and 2879 students¡¯ scores are tested by the probabilistic neural network algorithm, the nonlinear BP neural network algorithm and the perceptron neural network algorithm. By comparing these three kinds of neural network algorithms, we can get the following results: the train errors of these three neural network algorithms are all zero, but the test errors of these three neural network algorithms are different.and the test error of the probabilistic neural network algorithm is less than those of the nonlinear BP neural network algorithm and the perceptron algorithm; the train time of t!
he BP neural network algorithm is longer than those of the probabilistic neural network algorithm and the perceptron algorithm;the test time of the probabilistic neural network algorithm is longer that those of the nonlinear BP neural network algorithm and the perceptron algorithm. The correct rate of the probabilistic neural network algorithm heads to 99.06% when net.spread. The correct rate of the BP neural network algorithm changes from 98.51% to 99.06%. But the correct rate of the perceptron neural network algorithm is too low and changes from 20% to 30%. Therefore by considering the correct rate and the whole time of classification, we obtain that the probabilistic neural network algorithm is more suitable for solving the classification of the students¡¯ scores on only one course . And we take 1680 students' scores on five course to be analyzed and classified. Among these scored, 179 students' scores are trained and 1501 students¡¯ scores are tested by the nonlin!
ear BP neural network algorithm with the momentum factor, the nonlinea
r BP neural network algorithm with the gradient descent method, the probabilistic neural network algorithm and the self-organizing complete network algorithm. By comparing these kinds of neural network algorithms, we can get the following results: the train errors of the probabilistic neural network algorithm are all zero,those of the BP neural network algorithm with the momentum factor are all less than 0.0089, those of the BP neural network algorithm with the gradient descent method are all less than 0.0536, and those of the self-organizing compete neural network algorithm are all less than 0.4 and are all more than 0.2436; the test errors of the probabilistic neural network algorithm all equal to 0.0799, but those of the BP neural network algorithm with the momentum factor are all less than 0.0738, those of the BP neural network algorithm with the gradient descent method are all less than 0.1332, and those of the self-organizing compete neural network algorithm are !
all less than 0.3888 and are all more than 0.1871; the train times of the the probabilistic neural network algorithm are all less than 0.0469,those of the BP neural network algorithm with the momentum factor are all less than 33.0156 and are all more than 29.6875, those of the BP neural network algorithm with the gradient descent method are almost 24.3594 and are mostly less than 7.1875, and those of the self-organizing compete neural network algorithm are all less than 332.9219 and are all more than 310.0156; the test times of the probabilistic neural network algorithm are the least and are all less than 0.1719 and more than 1406, but those of the other neural network algorithms are all less than 0.0938; the train correct rates of the probabilistic neural network algorithm are all 100%when net.spread, those of the BP neural network algorithm with the momentum factor are all less than 99.44% and are all more than 97.77%, those of the BP neural network algorithm w!
ith the gradient descent method are all less than 93.30% and are all
more than 86.59%, and the those of the self-organizing compete neural network algorithm are all less than 40%;the test correct rates of the probabilistic neural network algorithm are all 80.01%, those of the BP neural network algorithm with the momentum factor are all less than 87.67% and are all more than 81.55%, those of the BP neural network algorithm with the gradient descent method are all less than 82.41% and are all more than 66.69%, and those of the self-organizing compete neural network algorithm are all less than 53.23%.Therefore by considering the correct rates and the whole times of classification, we obtain that the probabilistic neural network algorithm and the BP neural network algorithm are more suitable for solving the classification of the students¡¯ scores on five courses .
Keywords: The Students¡¯ Scores; BP Neural Network; Probabilistic Neural Network; Perceptron Neural network; Self-organizing Compete Neural Network; Train error; Test error; Train time; Test time;Train Correct Rate; Test Correct Rate
EXTENSION of the file: .doc
Special (Invited) Session: Classification of the Students Scores based on Neural Networks
Organizer of the Session: 102-294
How Did you learn about congress:
IP ADDRESS: 222.31.47.41
