Predicting Academic Performance of Deaf Students Using Feed Forward Neural Network and An Improved PSO Algorithm

• Published in: Webology Vol. 18; no. Special Issue 01; pp. 112 - 126
• Main Authors: Durga, V. Sathya, Jeyaprakash, Thangakumar
• Format: Journal Article
• Language: English
• Published: Tehran Dr. Alireza Noruzi, University of Tehran, Department of Library and Information Science 2021
• Subjects: Academic achievement; Accuracy; Adjustment; Algorithms; Classification; Datasets; Deafness; Decision trees; Literacy; Machine learning; Mean square errors; Neural networks; Optimization; Students; Support vector machines; India
• ISSN: 1735-188X; 1735-188X
• DOI: 10.14704/WEB/V18SI01/WEB18048

Literacy rate of deaf students is very less in India. So there is a need to build an effective academic prediction model for identifying weak deaf students. Many machine learning techniques like Decision tree, Support Vector Machine, Neural Network are used to build prediction models. But the most preferred technique is neural network. It is found out that regression model build with neural networks takes more time to converge and the error rate is quite high. To solve the problems of neural network, we use Particle Swarm Optimization (PSO) for weight adjustment in the neural network. But, one of the main drawback of PSO lies in setting the initial parameters. So, a new PSO algorithm which determines the initial weight of the neural network using regression equation is proposed. The results show that neural network build with the proposed PSO algorithm performs well than neural network build with basic PSO algorithm. The Mean Square Error (MSE) achieved in this work is 0.0998, which is comparatively less than many existing models.