Implementation of artificial neural network to assesment the lecturer's performance

2nd Nommensen International Conference on Technology and Engineering

Mulia Siregar, Victor Marudut^1 ; Sugara, Heru^1

Politeknik Bisnis Indonesia, Jl. Sriwijaya No. 9 C-E, Pematangsiantar Sumatera Utara, Indonesia^1

关键词: Activation functions;    Architectural modeling;    Error tolerance;    Hidden layers;    Learning rates;    Neural network backpropagation;    Training data;    Weight initialization;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/420/1/012112/pdf
DOI  :  10.1088/1757-899X/420/1/012112

来源: IOP

【 摘 要 】

The purpose of this research is to assess the performance of lecturers in teaching by using artificial neural network backpropagation. The tests were performed using Mathlab software that was tested with some forms of network architecture. The best architecture of artificial neural network (ANN) used is with architectural model 8-3-3-3-2. In the hidden layer used logsig activation function, and in the output layer used pureline activation function. In the first and second hidden layers using Nguyen Widrow weight initialization, the value of learning rate is0.1, error tolerance value is 0.001, with the maximum epoch is 3094 during training. With this 8-3-3-3-2 model ANN can recognize training data and test data up to 100% according to the desired target.